Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / dma / s3c24xx-dma.c
diff --git a/kernel/drivers/dma/s3c24xx-dma.c b/kernel/drivers/dma/s3c24xx-dma.c
new file mode 100644 (file)
index 0000000..01dcaf2
--- /dev/null
@@ -0,0 +1,1418 @@
+/*
+ * S3C24XX DMA handling
+ *
+ * Copyright (c) 2013 Heiko Stuebner <heiko@sntech.de>
+ *
+ * based on amba-pl08x.c
+ *
+ * Copyright (c) 2006 ARM Ltd.
+ * Copyright (c) 2010 ST-Ericsson SA
+ *
+ * Author: Peter Pearse <peter.pearse@arm.com>
+ * Author: Linus Walleij <linus.walleij@stericsson.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * The DMA controllers in S3C24XX SoCs have a varying number of DMA signals
+ * that can be routed to any of the 4 to 8 hardware-channels.
+ *
+ * Therefore on these DMA controllers the number of channels
+ * and the number of incoming DMA signals are two totally different things.
+ * It is usually not possible to theoretically handle all physical signals,
+ * so a multiplexing scheme with possible denial of use is necessary.
+ *
+ * Open items:
+ * - bursts
+ */
+
+#include <linux/platform_device.h>
+#include <linux/types.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
+#include <linux/clk.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/platform_data/dma-s3c24xx.h>
+
+#include "dmaengine.h"
+#include "virt-dma.h"
+
+#define MAX_DMA_CHANNELS       8
+
+#define S3C24XX_DISRC                  0x00
+#define S3C24XX_DISRCC                 0x04
+#define S3C24XX_DISRCC_INC_INCREMENT   0
+#define S3C24XX_DISRCC_INC_FIXED       BIT(0)
+#define S3C24XX_DISRCC_LOC_AHB         0
+#define S3C24XX_DISRCC_LOC_APB         BIT(1)
+
+#define S3C24XX_DIDST                  0x08
+#define S3C24XX_DIDSTC                 0x0c
+#define S3C24XX_DIDSTC_INC_INCREMENT   0
+#define S3C24XX_DIDSTC_INC_FIXED       BIT(0)
+#define S3C24XX_DIDSTC_LOC_AHB         0
+#define S3C24XX_DIDSTC_LOC_APB         BIT(1)
+#define S3C24XX_DIDSTC_INT_TC0         0
+#define S3C24XX_DIDSTC_INT_RELOAD      BIT(2)
+
+#define S3C24XX_DCON                   0x10
+
+#define S3C24XX_DCON_TC_MASK           0xfffff
+#define S3C24XX_DCON_DSZ_BYTE          (0 << 20)
+#define S3C24XX_DCON_DSZ_HALFWORD      (1 << 20)
+#define S3C24XX_DCON_DSZ_WORD          (2 << 20)
+#define S3C24XX_DCON_DSZ_MASK          (3 << 20)
+#define S3C24XX_DCON_DSZ_SHIFT         20
+#define S3C24XX_DCON_AUTORELOAD                0
+#define S3C24XX_DCON_NORELOAD          BIT(22)
+#define S3C24XX_DCON_HWTRIG            BIT(23)
+#define S3C24XX_DCON_HWSRC_SHIFT       24
+#define S3C24XX_DCON_SERV_SINGLE       0
+#define S3C24XX_DCON_SERV_WHOLE                BIT(27)
+#define S3C24XX_DCON_TSZ_UNIT          0
+#define S3C24XX_DCON_TSZ_BURST4                BIT(28)
+#define S3C24XX_DCON_INT               BIT(29)
+#define S3C24XX_DCON_SYNC_PCLK         0
+#define S3C24XX_DCON_SYNC_HCLK         BIT(30)
+#define S3C24XX_DCON_DEMAND            0
+#define S3C24XX_DCON_HANDSHAKE         BIT(31)
+
+#define S3C24XX_DSTAT                  0x14
+#define S3C24XX_DSTAT_STAT_BUSY                BIT(20)
+#define S3C24XX_DSTAT_CURRTC_MASK      0xfffff
+
+#define S3C24XX_DMASKTRIG              0x20
+#define S3C24XX_DMASKTRIG_SWTRIG       BIT(0)
+#define S3C24XX_DMASKTRIG_ON           BIT(1)
+#define S3C24XX_DMASKTRIG_STOP         BIT(2)
+
+#define S3C24XX_DMAREQSEL              0x24
+#define S3C24XX_DMAREQSEL_HW           BIT(0)
+
+/*
+ * S3C2410, S3C2440 and S3C2442 SoCs cannot select any physical channel
+ * for a DMA source. Instead only specific channels are valid.
+ * All of these SoCs have 4 physical channels and the number of request
+ * source bits is 3. Additionally we also need 1 bit to mark the channel
+ * as valid.
+ * Therefore we separate the chansel element of the channel data into 4
+ * parts of 4 bits each, to hold the information if the channel is valid
+ * and the hw request source to use.
+ *
+ * Example:
+ * SDI is valid on channels 0, 2 and 3 - with varying hw request sources.
+ * For it the chansel field would look like
+ *
+ * ((BIT(3) | 1) << 3 * 4) | // channel 3, with request source 1
+ * ((BIT(3) | 2) << 2 * 4) | // channel 2, with request source 2
+ * ((BIT(3) | 2) << 0 * 4)   // channel 0, with request source 2
+ */
+#define S3C24XX_CHANSEL_WIDTH          4
+#define S3C24XX_CHANSEL_VALID          BIT(3)
+#define S3C24XX_CHANSEL_REQ_MASK       7
+
+/*
+ * struct soc_data - vendor-specific config parameters for individual SoCs
+ * @stride: spacing between the registers of each channel
+ * @has_reqsel: does the controller use the newer requestselection mechanism
+ * @has_clocks: are controllable dma-clocks present
+ */
+struct soc_data {
+       int stride;
+       bool has_reqsel;
+       bool has_clocks;
+};
+
+/*
+ * enum s3c24xx_dma_chan_state - holds the virtual channel states
+ * @S3C24XX_DMA_CHAN_IDLE: the channel is idle
+ * @S3C24XX_DMA_CHAN_RUNNING: the channel has allocated a physical transport
+ * channel and is running a transfer on it
+ * @S3C24XX_DMA_CHAN_WAITING: the channel is waiting for a physical transport
+ * channel to become available (only pertains to memcpy channels)
+ */
+enum s3c24xx_dma_chan_state {
+       S3C24XX_DMA_CHAN_IDLE,
+       S3C24XX_DMA_CHAN_RUNNING,
+       S3C24XX_DMA_CHAN_WAITING,
+};
+
+/*
+ * struct s3c24xx_sg - structure containing data per sg
+ * @src_addr: src address of sg
+ * @dst_addr: dst address of sg
+ * @len: transfer len in bytes
+ * @node: node for txd's dsg_list
+ */
+struct s3c24xx_sg {
+       dma_addr_t src_addr;
+       dma_addr_t dst_addr;
+       size_t len;
+       struct list_head node;
+};
+
+/*
+ * struct s3c24xx_txd - wrapper for struct dma_async_tx_descriptor
+ * @vd: virtual DMA descriptor
+ * @dsg_list: list of children sg's
+ * @at: sg currently being transfered
+ * @width: transfer width
+ * @disrcc: value for source control register
+ * @didstc: value for destination control register
+ * @dcon: base value for dcon register
+ * @cyclic: indicate cyclic transfer
+ */
+struct s3c24xx_txd {
+       struct virt_dma_desc vd;
+       struct list_head dsg_list;
+       struct list_head *at;
+       u8 width;
+       u32 disrcc;
+       u32 didstc;
+       u32 dcon;
+       bool cyclic;
+};
+
+struct s3c24xx_dma_chan;
+
+/*
+ * struct s3c24xx_dma_phy - holder for the physical channels
+ * @id: physical index to this channel
+ * @valid: does the channel have all required elements
+ * @base: virtual memory base (remapped) for the this channel
+ * @irq: interrupt for this channel
+ * @clk: clock for this channel
+ * @lock: a lock to use when altering an instance of this struct
+ * @serving: virtual channel currently being served by this physicalchannel
+ * @host: a pointer to the host (internal use)
+ */
+struct s3c24xx_dma_phy {
+       unsigned int                    id;
+       bool                            valid;
+       void __iomem                    *base;
+       int                             irq;
+       struct clk                      *clk;
+       spinlock_t                      lock;
+       struct s3c24xx_dma_chan         *serving;
+       struct s3c24xx_dma_engine       *host;
+};
+
+/*
+ * struct s3c24xx_dma_chan - this structure wraps a DMA ENGINE channel
+ * @id: the id of the channel
+ * @name: name of the channel
+ * @vc: wrappped virtual channel
+ * @phy: the physical channel utilized by this channel, if there is one
+ * @runtime_addr: address for RX/TX according to the runtime config
+ * @at: active transaction on this channel
+ * @lock: a lock for this channel data
+ * @host: a pointer to the host (internal use)
+ * @state: whether the channel is idle, running etc
+ * @slave: whether this channel is a device (slave) or for memcpy
+ */
+struct s3c24xx_dma_chan {
+       int id;
+       const char *name;
+       struct virt_dma_chan vc;
+       struct s3c24xx_dma_phy *phy;
+       struct dma_slave_config cfg;
+       struct s3c24xx_txd *at;
+       struct s3c24xx_dma_engine *host;
+       enum s3c24xx_dma_chan_state state;
+       bool slave;
+};
+
+/*
+ * struct s3c24xx_dma_engine - the local state holder for the S3C24XX
+ * @pdev: the corresponding platform device
+ * @pdata: platform data passed in from the platform/machine
+ * @base: virtual memory base (remapped)
+ * @slave: slave engine for this instance
+ * @memcpy: memcpy engine for this instance
+ * @phy_chans: array of data for the physical channels
+ */
+struct s3c24xx_dma_engine {
+       struct platform_device                  *pdev;
+       const struct s3c24xx_dma_platdata       *pdata;
+       struct soc_data                         *sdata;
+       void __iomem                            *base;
+       struct dma_device                       slave;
+       struct dma_device                       memcpy;
+       struct s3c24xx_dma_phy                  *phy_chans;
+};
+
+/*
+ * Physical channel handling
+ */
+
+/*
+ * Check whether a certain channel is busy or not.
+ */
+static int s3c24xx_dma_phy_busy(struct s3c24xx_dma_phy *phy)
+{
+       unsigned int val = readl(phy->base + S3C24XX_DSTAT);
+       return val & S3C24XX_DSTAT_STAT_BUSY;
+}
+
+static bool s3c24xx_dma_phy_valid(struct s3c24xx_dma_chan *s3cchan,
+                                 struct s3c24xx_dma_phy *phy)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
+       struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
+       int phyvalid;
+
+       /* every phy is valid for memcopy channels */
+       if (!s3cchan->slave)
+               return true;
+
+       /* On newer variants all phys can be used for all virtual channels */
+       if (s3cdma->sdata->has_reqsel)
+               return true;
+
+       phyvalid = (cdata->chansel >> (phy->id * S3C24XX_CHANSEL_WIDTH));
+       return (phyvalid & S3C24XX_CHANSEL_VALID) ? true : false;
+}
+
+/*
+ * Allocate a physical channel for a virtual channel
+ *
+ * Try to locate a physical channel to be used for this transfer. If all
+ * are taken return NULL and the requester will have to cope by using
+ * some fallback PIO mode or retrying later.
+ */
+static
+struct s3c24xx_dma_phy *s3c24xx_dma_get_phy(struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
+       struct s3c24xx_dma_channel *cdata;
+       struct s3c24xx_dma_phy *phy = NULL;
+       unsigned long flags;
+       int i;
+       int ret;
+
+       if (s3cchan->slave)
+               cdata = &pdata->channels[s3cchan->id];
+
+       for (i = 0; i < s3cdma->pdata->num_phy_channels; i++) {
+               phy = &s3cdma->phy_chans[i];
+
+               if (!phy->valid)
+                       continue;
+
+               if (!s3c24xx_dma_phy_valid(s3cchan, phy))
+                       continue;
+
+               spin_lock_irqsave(&phy->lock, flags);
+
+               if (!phy->serving) {
+                       phy->serving = s3cchan;
+                       spin_unlock_irqrestore(&phy->lock, flags);
+                       break;
+               }
+
+               spin_unlock_irqrestore(&phy->lock, flags);
+       }
+
+       /* No physical channel available, cope with it */
+       if (i == s3cdma->pdata->num_phy_channels) {
+               dev_warn(&s3cdma->pdev->dev, "no phy channel available\n");
+               return NULL;
+       }
+
+       /* start the phy clock */
+       if (s3cdma->sdata->has_clocks) {
+               ret = clk_enable(phy->clk);
+               if (ret) {
+                       dev_err(&s3cdma->pdev->dev, "could not enable clock for channel %d, err %d\n",
+                               phy->id, ret);
+                       phy->serving = NULL;
+                       return NULL;
+               }
+       }
+
+       return phy;
+}
+
+/*
+ * Mark the physical channel as free.
+ *
+ * This drops the link between the physical and virtual channel.
+ */
+static inline void s3c24xx_dma_put_phy(struct s3c24xx_dma_phy *phy)
+{
+       struct s3c24xx_dma_engine *s3cdma = phy->host;
+
+       if (s3cdma->sdata->has_clocks)
+               clk_disable(phy->clk);
+
+       phy->serving = NULL;
+}
+
+/*
+ * Stops the channel by writing the stop bit.
+ * This should not be used for an on-going transfer, but as a method of
+ * shutting down a channel (eg, when it's no longer used) or terminating a
+ * transfer.
+ */
+static void s3c24xx_dma_terminate_phy(struct s3c24xx_dma_phy *phy)
+{
+       writel(S3C24XX_DMASKTRIG_STOP, phy->base + S3C24XX_DMASKTRIG);
+}
+
+/*
+ * Virtual channel handling
+ */
+
+static inline
+struct s3c24xx_dma_chan *to_s3c24xx_dma_chan(struct dma_chan *chan)
+{
+       return container_of(chan, struct s3c24xx_dma_chan, vc.chan);
+}
+
+static u32 s3c24xx_dma_getbytes_chan(struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_phy *phy = s3cchan->phy;
+       struct s3c24xx_txd *txd = s3cchan->at;
+       u32 tc = readl(phy->base + S3C24XX_DSTAT) & S3C24XX_DSTAT_CURRTC_MASK;
+
+       return tc * txd->width;
+}
+
+static int s3c24xx_dma_set_runtime_config(struct dma_chan *chan,
+                                 struct dma_slave_config *config)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       unsigned long flags;
+       int ret = 0;
+
+       /* Reject definitely invalid configurations */
+       if (config->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
+           config->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
+               return -EINVAL;
+
+       spin_lock_irqsave(&s3cchan->vc.lock, flags);
+
+       if (!s3cchan->slave) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       s3cchan->cfg = *config;
+
+out:
+       spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+       return ret;
+}
+
+/*
+ * Transfer handling
+ */
+
+static inline
+struct s3c24xx_txd *to_s3c24xx_txd(struct dma_async_tx_descriptor *tx)
+{
+       return container_of(tx, struct s3c24xx_txd, vd.tx);
+}
+
+static struct s3c24xx_txd *s3c24xx_dma_get_txd(void)
+{
+       struct s3c24xx_txd *txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
+
+       if (txd) {
+               INIT_LIST_HEAD(&txd->dsg_list);
+               txd->dcon = S3C24XX_DCON_INT | S3C24XX_DCON_NORELOAD;
+       }
+
+       return txd;
+}
+
+static void s3c24xx_dma_free_txd(struct s3c24xx_txd *txd)
+{
+       struct s3c24xx_sg *dsg, *_dsg;
+
+       list_for_each_entry_safe(dsg, _dsg, &txd->dsg_list, node) {
+               list_del(&dsg->node);
+               kfree(dsg);
+       }
+
+       kfree(txd);
+}
+
+static void s3c24xx_dma_start_next_sg(struct s3c24xx_dma_chan *s3cchan,
+                                      struct s3c24xx_txd *txd)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       struct s3c24xx_dma_phy *phy = s3cchan->phy;
+       const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
+       struct s3c24xx_sg *dsg = list_entry(txd->at, struct s3c24xx_sg, node);
+       u32 dcon = txd->dcon;
+       u32 val;
+
+       /* transfer-size and -count from len and width */
+       switch (txd->width) {
+       case 1:
+               dcon |= S3C24XX_DCON_DSZ_BYTE | dsg->len;
+               break;
+       case 2:
+               dcon |= S3C24XX_DCON_DSZ_HALFWORD | (dsg->len / 2);
+               break;
+       case 4:
+               dcon |= S3C24XX_DCON_DSZ_WORD | (dsg->len / 4);
+               break;
+       }
+
+       if (s3cchan->slave) {
+               struct s3c24xx_dma_channel *cdata =
+                                       &pdata->channels[s3cchan->id];
+
+               if (s3cdma->sdata->has_reqsel) {
+                       writel_relaxed((cdata->chansel << 1) |
+                                                       S3C24XX_DMAREQSEL_HW,
+                                       phy->base + S3C24XX_DMAREQSEL);
+               } else {
+                       int csel = cdata->chansel >> (phy->id *
+                                                       S3C24XX_CHANSEL_WIDTH);
+
+                       csel &= S3C24XX_CHANSEL_REQ_MASK;
+                       dcon |= csel << S3C24XX_DCON_HWSRC_SHIFT;
+                       dcon |= S3C24XX_DCON_HWTRIG;
+               }
+       } else {
+               if (s3cdma->sdata->has_reqsel)
+                       writel_relaxed(0, phy->base + S3C24XX_DMAREQSEL);
+       }
+
+       writel_relaxed(dsg->src_addr, phy->base + S3C24XX_DISRC);
+       writel_relaxed(txd->disrcc, phy->base + S3C24XX_DISRCC);
+       writel_relaxed(dsg->dst_addr, phy->base + S3C24XX_DIDST);
+       writel_relaxed(txd->didstc, phy->base + S3C24XX_DIDSTC);
+       writel_relaxed(dcon, phy->base + S3C24XX_DCON);
+
+       val = readl_relaxed(phy->base + S3C24XX_DMASKTRIG);
+       val &= ~S3C24XX_DMASKTRIG_STOP;
+       val |= S3C24XX_DMASKTRIG_ON;
+
+       /* trigger the dma operation for memcpy transfers */
+       if (!s3cchan->slave)
+               val |= S3C24XX_DMASKTRIG_SWTRIG;
+
+       writel(val, phy->base + S3C24XX_DMASKTRIG);
+}
+
+/*
+ * Set the initial DMA register values and start first sg.
+ */
+static void s3c24xx_dma_start_next_txd(struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_phy *phy = s3cchan->phy;
+       struct virt_dma_desc *vd = vchan_next_desc(&s3cchan->vc);
+       struct s3c24xx_txd *txd = to_s3c24xx_txd(&vd->tx);
+
+       list_del(&txd->vd.node);
+
+       s3cchan->at = txd;
+
+       /* Wait for channel inactive */
+       while (s3c24xx_dma_phy_busy(phy))
+               cpu_relax();
+
+       /* point to the first element of the sg list */
+       txd->at = txd->dsg_list.next;
+       s3c24xx_dma_start_next_sg(s3cchan, txd);
+}
+
+static void s3c24xx_dma_free_txd_list(struct s3c24xx_dma_engine *s3cdma,
+                               struct s3c24xx_dma_chan *s3cchan)
+{
+       LIST_HEAD(head);
+
+       vchan_get_all_descriptors(&s3cchan->vc, &head);
+       vchan_dma_desc_free_list(&s3cchan->vc, &head);
+}
+
+/*
+ * Try to allocate a physical channel.  When successful, assign it to
+ * this virtual channel, and initiate the next descriptor.  The
+ * virtual channel lock must be held at this point.
+ */
+static void s3c24xx_dma_phy_alloc_and_start(struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       struct s3c24xx_dma_phy *phy;
+
+       phy = s3c24xx_dma_get_phy(s3cchan);
+       if (!phy) {
+               dev_dbg(&s3cdma->pdev->dev, "no physical channel available for xfer on %s\n",
+                       s3cchan->name);
+               s3cchan->state = S3C24XX_DMA_CHAN_WAITING;
+               return;
+       }
+
+       dev_dbg(&s3cdma->pdev->dev, "allocated physical channel %d for xfer on %s\n",
+               phy->id, s3cchan->name);
+
+       s3cchan->phy = phy;
+       s3cchan->state = S3C24XX_DMA_CHAN_RUNNING;
+
+       s3c24xx_dma_start_next_txd(s3cchan);
+}
+
+static void s3c24xx_dma_phy_reassign_start(struct s3c24xx_dma_phy *phy,
+       struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+
+       dev_dbg(&s3cdma->pdev->dev, "reassigned physical channel %d for xfer on %s\n",
+               phy->id, s3cchan->name);
+
+       /*
+        * We do this without taking the lock; we're really only concerned
+        * about whether this pointer is NULL or not, and we're guaranteed
+        * that this will only be called when it _already_ is non-NULL.
+        */
+       phy->serving = s3cchan;
+       s3cchan->phy = phy;
+       s3cchan->state = S3C24XX_DMA_CHAN_RUNNING;
+       s3c24xx_dma_start_next_txd(s3cchan);
+}
+
+/*
+ * Free a physical DMA channel, potentially reallocating it to another
+ * virtual channel if we have any pending.
+ */
+static void s3c24xx_dma_phy_free(struct s3c24xx_dma_chan *s3cchan)
+{
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       struct s3c24xx_dma_chan *p, *next;
+
+retry:
+       next = NULL;
+
+       /* Find a waiting virtual channel for the next transfer. */
+       list_for_each_entry(p, &s3cdma->memcpy.channels, vc.chan.device_node)
+               if (p->state == S3C24XX_DMA_CHAN_WAITING) {
+                       next = p;
+                       break;
+               }
+
+       if (!next) {
+               list_for_each_entry(p, &s3cdma->slave.channels,
+                                   vc.chan.device_node)
+                       if (p->state == S3C24XX_DMA_CHAN_WAITING &&
+                                     s3c24xx_dma_phy_valid(p, s3cchan->phy)) {
+                               next = p;
+                               break;
+                       }
+       }
+
+       /* Ensure that the physical channel is stopped */
+       s3c24xx_dma_terminate_phy(s3cchan->phy);
+
+       if (next) {
+               bool success;
+
+               /*
+                * Eww.  We know this isn't going to deadlock
+                * but lockdep probably doesn't.
+                */
+               spin_lock(&next->vc.lock);
+               /* Re-check the state now that we have the lock */
+               success = next->state == S3C24XX_DMA_CHAN_WAITING;
+               if (success)
+                       s3c24xx_dma_phy_reassign_start(s3cchan->phy, next);
+               spin_unlock(&next->vc.lock);
+
+               /* If the state changed, try to find another channel */
+               if (!success)
+                       goto retry;
+       } else {
+               /* No more jobs, so free up the physical channel */
+               s3c24xx_dma_put_phy(s3cchan->phy);
+       }
+
+       s3cchan->phy = NULL;
+       s3cchan->state = S3C24XX_DMA_CHAN_IDLE;
+}
+
+static void s3c24xx_dma_desc_free(struct virt_dma_desc *vd)
+{
+       struct s3c24xx_txd *txd = to_s3c24xx_txd(&vd->tx);
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(vd->tx.chan);
+
+       if (!s3cchan->slave)
+               dma_descriptor_unmap(&vd->tx);
+
+       s3c24xx_dma_free_txd(txd);
+}
+
+static irqreturn_t s3c24xx_dma_irq(int irq, void *data)
+{
+       struct s3c24xx_dma_phy *phy = data;
+       struct s3c24xx_dma_chan *s3cchan = phy->serving;
+       struct s3c24xx_txd *txd;
+
+       dev_dbg(&phy->host->pdev->dev, "interrupt on channel %d\n", phy->id);
+
+       /*
+        * Interrupts happen to notify the completion of a transfer and the
+        * channel should have moved into its stop state already on its own.
+        * Therefore interrupts on channels not bound to a virtual channel
+        * should never happen. Nevertheless send a terminate command to the
+        * channel if the unlikely case happens.
+        */
+       if (unlikely(!s3cchan)) {
+               dev_err(&phy->host->pdev->dev, "interrupt on unused channel %d\n",
+                       phy->id);
+
+               s3c24xx_dma_terminate_phy(phy);
+
+               return IRQ_HANDLED;
+       }
+
+       spin_lock(&s3cchan->vc.lock);
+       txd = s3cchan->at;
+       if (txd) {
+               /* when more sg's are in this txd, start the next one */
+               if (!list_is_last(txd->at, &txd->dsg_list)) {
+                       txd->at = txd->at->next;
+                       if (txd->cyclic)
+                               vchan_cyclic_callback(&txd->vd);
+                       s3c24xx_dma_start_next_sg(s3cchan, txd);
+               } else if (!txd->cyclic) {
+                       s3cchan->at = NULL;
+                       vchan_cookie_complete(&txd->vd);
+
+                       /*
+                        * And start the next descriptor (if any),
+                        * otherwise free this channel.
+                        */
+                       if (vchan_next_desc(&s3cchan->vc))
+                               s3c24xx_dma_start_next_txd(s3cchan);
+                       else
+                               s3c24xx_dma_phy_free(s3cchan);
+               } else {
+                       vchan_cyclic_callback(&txd->vd);
+
+                       /* Cyclic: reset at beginning */
+                       txd->at = txd->dsg_list.next;
+                       s3c24xx_dma_start_next_sg(s3cchan, txd);
+               }
+       }
+       spin_unlock(&s3cchan->vc.lock);
+
+       return IRQ_HANDLED;
+}
+
+/*
+ * The DMA ENGINE API
+ */
+
+static int s3c24xx_dma_terminate_all(struct dma_chan *chan)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       unsigned long flags;
+       int ret = 0;
+
+       spin_lock_irqsave(&s3cchan->vc.lock, flags);
+
+       if (!s3cchan->phy && !s3cchan->at) {
+               dev_err(&s3cdma->pdev->dev, "trying to terminate already stopped channel %d\n",
+                       s3cchan->id);
+               ret = -EINVAL;
+               goto unlock;
+       }
+
+       s3cchan->state = S3C24XX_DMA_CHAN_IDLE;
+
+       /* Mark physical channel as free */
+       if (s3cchan->phy)
+               s3c24xx_dma_phy_free(s3cchan);
+
+       /* Dequeue current job */
+       if (s3cchan->at) {
+               s3c24xx_dma_desc_free(&s3cchan->at->vd);
+               s3cchan->at = NULL;
+       }
+
+       /* Dequeue jobs not yet fired as well */
+       s3c24xx_dma_free_txd_list(s3cdma, s3cchan);
+unlock:
+       spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+
+       return ret;
+}
+
+static void s3c24xx_dma_free_chan_resources(struct dma_chan *chan)
+{
+       /* Ensure all queued descriptors are freed */
+       vchan_free_chan_resources(to_virt_chan(chan));
+}
+
+static enum dma_status s3c24xx_dma_tx_status(struct dma_chan *chan,
+               dma_cookie_t cookie, struct dma_tx_state *txstate)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       struct s3c24xx_txd *txd;
+       struct s3c24xx_sg *dsg;
+       struct virt_dma_desc *vd;
+       unsigned long flags;
+       enum dma_status ret;
+       size_t bytes = 0;
+
+       spin_lock_irqsave(&s3cchan->vc.lock, flags);
+       ret = dma_cookie_status(chan, cookie, txstate);
+       if (ret == DMA_COMPLETE) {
+               spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+               return ret;
+       }
+
+       /*
+        * There's no point calculating the residue if there's
+        * no txstate to store the value.
+        */
+       if (!txstate) {
+               spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+               return ret;
+       }
+
+       vd = vchan_find_desc(&s3cchan->vc, cookie);
+       if (vd) {
+               /* On the issued list, so hasn't been processed yet */
+               txd = to_s3c24xx_txd(&vd->tx);
+
+               list_for_each_entry(dsg, &txd->dsg_list, node)
+                       bytes += dsg->len;
+       } else {
+               /*
+                * Currently running, so sum over the pending sg's and
+                * the currently active one.
+                */
+               txd = s3cchan->at;
+
+               dsg = list_entry(txd->at, struct s3c24xx_sg, node);
+               list_for_each_entry_from(dsg, &txd->dsg_list, node)
+                       bytes += dsg->len;
+
+               bytes += s3c24xx_dma_getbytes_chan(s3cchan);
+       }
+       spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+
+       /*
+        * This cookie not complete yet
+        * Get number of bytes left in the active transactions and queue
+        */
+       dma_set_residue(txstate, bytes);
+
+       /* Whether waiting or running, we're in progress */
+       return ret;
+}
+
+/*
+ * Initialize a descriptor to be used by memcpy submit
+ */
+static struct dma_async_tx_descriptor *s3c24xx_dma_prep_memcpy(
+               struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
+               size_t len, unsigned long flags)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       struct s3c24xx_txd *txd;
+       struct s3c24xx_sg *dsg;
+       int src_mod, dest_mod;
+
+       dev_dbg(&s3cdma->pdev->dev, "prepare memcpy of %d bytes from %s\n",
+                       len, s3cchan->name);
+
+       if ((len & S3C24XX_DCON_TC_MASK) != len) {
+               dev_err(&s3cdma->pdev->dev, "memcpy size %d to large\n", len);
+               return NULL;
+       }
+
+       txd = s3c24xx_dma_get_txd();
+       if (!txd)
+               return NULL;
+
+       dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
+       if (!dsg) {
+               s3c24xx_dma_free_txd(txd);
+               return NULL;
+       }
+       list_add_tail(&dsg->node, &txd->dsg_list);
+
+       dsg->src_addr = src;
+       dsg->dst_addr = dest;
+       dsg->len = len;
+
+       /*
+        * Determine a suitable transfer width.
+        * The DMA controller cannot fetch/store information which is not
+        * naturally aligned on the bus, i.e., a 4 byte fetch must start at
+        * an address divisible by 4 - more generally addr % width must be 0.
+        */
+       src_mod = src % 4;
+       dest_mod = dest % 4;
+       switch (len % 4) {
+       case 0:
+               txd->width = (src_mod == 0 && dest_mod == 0) ? 4 : 1;
+               break;
+       case 2:
+               txd->width = ((src_mod == 2 || src_mod == 0) &&
+                             (dest_mod == 2 || dest_mod == 0)) ? 2 : 1;
+               break;
+       default:
+               txd->width = 1;
+               break;
+       }
+
+       txd->disrcc = S3C24XX_DISRCC_LOC_AHB | S3C24XX_DISRCC_INC_INCREMENT;
+       txd->didstc = S3C24XX_DIDSTC_LOC_AHB | S3C24XX_DIDSTC_INC_INCREMENT;
+       txd->dcon |= S3C24XX_DCON_DEMAND | S3C24XX_DCON_SYNC_HCLK |
+                    S3C24XX_DCON_SERV_WHOLE;
+
+       return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
+}
+
+static struct dma_async_tx_descriptor *s3c24xx_dma_prep_dma_cyclic(
+       struct dma_chan *chan, dma_addr_t addr, size_t size, size_t period,
+       enum dma_transfer_direction direction, unsigned long flags)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
+       struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
+       struct s3c24xx_txd *txd;
+       struct s3c24xx_sg *dsg;
+       unsigned sg_len;
+       dma_addr_t slave_addr;
+       u32 hwcfg = 0;
+       int i;
+
+       dev_dbg(&s3cdma->pdev->dev,
+               "prepare cyclic transaction of %zu bytes with period %zu from %s\n",
+               size, period, s3cchan->name);
+
+       if (!is_slave_direction(direction)) {
+               dev_err(&s3cdma->pdev->dev,
+                       "direction %d unsupported\n", direction);
+               return NULL;
+       }
+
+       txd = s3c24xx_dma_get_txd();
+       if (!txd)
+               return NULL;
+
+       txd->cyclic = 1;
+
+       if (cdata->handshake)
+               txd->dcon |= S3C24XX_DCON_HANDSHAKE;
+
+       switch (cdata->bus) {
+       case S3C24XX_DMA_APB:
+               txd->dcon |= S3C24XX_DCON_SYNC_PCLK;
+               hwcfg |= S3C24XX_DISRCC_LOC_APB;
+               break;
+       case S3C24XX_DMA_AHB:
+               txd->dcon |= S3C24XX_DCON_SYNC_HCLK;
+               hwcfg |= S3C24XX_DISRCC_LOC_AHB;
+               break;
+       }
+
+       /*
+        * Always assume our peripheral desintation is a fixed
+        * address in memory.
+        */
+       hwcfg |= S3C24XX_DISRCC_INC_FIXED;
+
+       /*
+        * Individual dma operations are requested by the slave,
+        * so serve only single atomic operations (S3C24XX_DCON_SERV_SINGLE).
+        */
+       txd->dcon |= S3C24XX_DCON_SERV_SINGLE;
+
+       if (direction == DMA_MEM_TO_DEV) {
+               txd->disrcc = S3C24XX_DISRCC_LOC_AHB |
+                             S3C24XX_DISRCC_INC_INCREMENT;
+               txd->didstc = hwcfg;
+               slave_addr = s3cchan->cfg.dst_addr;
+               txd->width = s3cchan->cfg.dst_addr_width;
+       } else {
+               txd->disrcc = hwcfg;
+               txd->didstc = S3C24XX_DIDSTC_LOC_AHB |
+                             S3C24XX_DIDSTC_INC_INCREMENT;
+               slave_addr = s3cchan->cfg.src_addr;
+               txd->width = s3cchan->cfg.src_addr_width;
+       }
+
+       sg_len = size / period;
+
+       for (i = 0; i < sg_len; i++) {
+               dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
+               if (!dsg) {
+                       s3c24xx_dma_free_txd(txd);
+                       return NULL;
+               }
+               list_add_tail(&dsg->node, &txd->dsg_list);
+
+               dsg->len = period;
+               /* Check last period length */
+               if (i == sg_len - 1)
+                       dsg->len = size - period * i;
+               if (direction == DMA_MEM_TO_DEV) {
+                       dsg->src_addr = addr + period * i;
+                       dsg->dst_addr = slave_addr;
+               } else { /* DMA_DEV_TO_MEM */
+                       dsg->src_addr = slave_addr;
+                       dsg->dst_addr = addr + period * i;
+               }
+       }
+
+       return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
+}
+
+static struct dma_async_tx_descriptor *s3c24xx_dma_prep_slave_sg(
+               struct dma_chan *chan, struct scatterlist *sgl,
+               unsigned int sg_len, enum dma_transfer_direction direction,
+               unsigned long flags, void *context)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
+       const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
+       struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
+       struct s3c24xx_txd *txd;
+       struct s3c24xx_sg *dsg;
+       struct scatterlist *sg;
+       dma_addr_t slave_addr;
+       u32 hwcfg = 0;
+       int tmp;
+
+       dev_dbg(&s3cdma->pdev->dev, "prepare transaction of %d bytes from %s\n",
+                       sg_dma_len(sgl), s3cchan->name);
+
+       txd = s3c24xx_dma_get_txd();
+       if (!txd)
+               return NULL;
+
+       if (cdata->handshake)
+               txd->dcon |= S3C24XX_DCON_HANDSHAKE;
+
+       switch (cdata->bus) {
+       case S3C24XX_DMA_APB:
+               txd->dcon |= S3C24XX_DCON_SYNC_PCLK;
+               hwcfg |= S3C24XX_DISRCC_LOC_APB;
+               break;
+       case S3C24XX_DMA_AHB:
+               txd->dcon |= S3C24XX_DCON_SYNC_HCLK;
+               hwcfg |= S3C24XX_DISRCC_LOC_AHB;
+               break;
+       }
+
+       /*
+        * Always assume our peripheral desintation is a fixed
+        * address in memory.
+        */
+       hwcfg |= S3C24XX_DISRCC_INC_FIXED;
+
+       /*
+        * Individual dma operations are requested by the slave,
+        * so serve only single atomic operations (S3C24XX_DCON_SERV_SINGLE).
+        */
+       txd->dcon |= S3C24XX_DCON_SERV_SINGLE;
+
+       if (direction == DMA_MEM_TO_DEV) {
+               txd->disrcc = S3C24XX_DISRCC_LOC_AHB |
+                             S3C24XX_DISRCC_INC_INCREMENT;
+               txd->didstc = hwcfg;
+               slave_addr = s3cchan->cfg.dst_addr;
+               txd->width = s3cchan->cfg.dst_addr_width;
+       } else if (direction == DMA_DEV_TO_MEM) {
+               txd->disrcc = hwcfg;
+               txd->didstc = S3C24XX_DIDSTC_LOC_AHB |
+                             S3C24XX_DIDSTC_INC_INCREMENT;
+               slave_addr = s3cchan->cfg.src_addr;
+               txd->width = s3cchan->cfg.src_addr_width;
+       } else {
+               s3c24xx_dma_free_txd(txd);
+               dev_err(&s3cdma->pdev->dev,
+                       "direction %d unsupported\n", direction);
+               return NULL;
+       }
+
+       for_each_sg(sgl, sg, sg_len, tmp) {
+               dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
+               if (!dsg) {
+                       s3c24xx_dma_free_txd(txd);
+                       return NULL;
+               }
+               list_add_tail(&dsg->node, &txd->dsg_list);
+
+               dsg->len = sg_dma_len(sg);
+               if (direction == DMA_MEM_TO_DEV) {
+                       dsg->src_addr = sg_dma_address(sg);
+                       dsg->dst_addr = slave_addr;
+               } else { /* DMA_DEV_TO_MEM */
+                       dsg->src_addr = slave_addr;
+                       dsg->dst_addr = sg_dma_address(sg);
+               }
+       }
+
+       return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
+}
+
+/*
+ * Slave transactions callback to the slave device to allow
+ * synchronization of slave DMA signals with the DMAC enable
+ */
+static void s3c24xx_dma_issue_pending(struct dma_chan *chan)
+{
+       struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
+       unsigned long flags;
+
+       spin_lock_irqsave(&s3cchan->vc.lock, flags);
+       if (vchan_issue_pending(&s3cchan->vc)) {
+               if (!s3cchan->phy && s3cchan->state != S3C24XX_DMA_CHAN_WAITING)
+                       s3c24xx_dma_phy_alloc_and_start(s3cchan);
+       }
+       spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
+}
+
+/*
+ * Bringup and teardown
+ */
+
+/*
+ * Initialise the DMAC memcpy/slave channels.
+ * Make a local wrapper to hold required data
+ */
+static int s3c24xx_dma_init_virtual_channels(struct s3c24xx_dma_engine *s3cdma,
+               struct dma_device *dmadev, unsigned int channels, bool slave)
+{
+       struct s3c24xx_dma_chan *chan;
+       int i;
+
+       INIT_LIST_HEAD(&dmadev->channels);
+
+       /*
+        * Register as many many memcpy as we have physical channels,
+        * we won't always be able to use all but the code will have
+        * to cope with that situation.
+        */
+       for (i = 0; i < channels; i++) {
+               chan = devm_kzalloc(dmadev->dev, sizeof(*chan), GFP_KERNEL);
+               if (!chan) {
+                       dev_err(dmadev->dev,
+                               "%s no memory for channel\n", __func__);
+                       return -ENOMEM;
+               }
+
+               chan->id = i;
+               chan->host = s3cdma;
+               chan->state = S3C24XX_DMA_CHAN_IDLE;
+
+               if (slave) {
+                       chan->slave = true;
+                       chan->name = kasprintf(GFP_KERNEL, "slave%d", i);
+                       if (!chan->name)
+                               return -ENOMEM;
+               } else {
+                       chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
+                       if (!chan->name)
+                               return -ENOMEM;
+               }
+               dev_dbg(dmadev->dev,
+                        "initialize virtual channel \"%s\"\n",
+                        chan->name);
+
+               chan->vc.desc_free = s3c24xx_dma_desc_free;
+               vchan_init(&chan->vc, dmadev);
+       }
+       dev_info(dmadev->dev, "initialized %d virtual %s channels\n",
+                i, slave ? "slave" : "memcpy");
+       return i;
+}
+
+static void s3c24xx_dma_free_virtual_channels(struct dma_device *dmadev)
+{
+       struct s3c24xx_dma_chan *chan = NULL;
+       struct s3c24xx_dma_chan *next;
+
+       list_for_each_entry_safe(chan,
+                                next, &dmadev->channels, vc.chan.device_node)
+               list_del(&chan->vc.chan.device_node);
+}
+
+/* s3c2410, s3c2440 and s3c2442 have a 0x40 stride without separate clocks */
+static struct soc_data soc_s3c2410 = {
+       .stride = 0x40,
+       .has_reqsel = false,
+       .has_clocks = false,
+};
+
+/* s3c2412 and s3c2413 have a 0x40 stride and dmareqsel mechanism */
+static struct soc_data soc_s3c2412 = {
+       .stride = 0x40,
+       .has_reqsel = true,
+       .has_clocks = true,
+};
+
+/* s3c2443 and following have a 0x100 stride and dmareqsel mechanism */
+static struct soc_data soc_s3c2443 = {
+       .stride = 0x100,
+       .has_reqsel = true,
+       .has_clocks = true,
+};
+
+static struct platform_device_id s3c24xx_dma_driver_ids[] = {
+       {
+               .name           = "s3c2410-dma",
+               .driver_data    = (kernel_ulong_t)&soc_s3c2410,
+       }, {
+               .name           = "s3c2412-dma",
+               .driver_data    = (kernel_ulong_t)&soc_s3c2412,
+       }, {
+               .name           = "s3c2443-dma",
+               .driver_data    = (kernel_ulong_t)&soc_s3c2443,
+       },
+       { },
+};
+
+static struct soc_data *s3c24xx_dma_get_soc_data(struct platform_device *pdev)
+{
+       return (struct soc_data *)
+                        platform_get_device_id(pdev)->driver_data;
+}
+
+static int s3c24xx_dma_probe(struct platform_device *pdev)
+{
+       const struct s3c24xx_dma_platdata *pdata = dev_get_platdata(&pdev->dev);
+       struct s3c24xx_dma_engine *s3cdma;
+       struct soc_data *sdata;
+       struct resource *res;
+       int ret;
+       int i;
+
+       if (!pdata) {
+               dev_err(&pdev->dev, "platform data missing\n");
+               return -ENODEV;
+       }
+
+       /* Basic sanity check */
+       if (pdata->num_phy_channels > MAX_DMA_CHANNELS) {
+               dev_err(&pdev->dev, "to many dma channels %d, max %d\n",
+                       pdata->num_phy_channels, MAX_DMA_CHANNELS);
+               return -EINVAL;
+       }
+
+       sdata = s3c24xx_dma_get_soc_data(pdev);
+       if (!sdata)
+               return -EINVAL;
+
+       s3cdma = devm_kzalloc(&pdev->dev, sizeof(*s3cdma), GFP_KERNEL);
+       if (!s3cdma)
+               return -ENOMEM;
+
+       s3cdma->pdev = pdev;
+       s3cdma->pdata = pdata;
+       s3cdma->sdata = sdata;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       s3cdma->base = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(s3cdma->base))
+               return PTR_ERR(s3cdma->base);
+
+       s3cdma->phy_chans = devm_kzalloc(&pdev->dev,
+                                             sizeof(struct s3c24xx_dma_phy) *
+                                                       pdata->num_phy_channels,
+                                             GFP_KERNEL);
+       if (!s3cdma->phy_chans)
+               return -ENOMEM;
+
+       /* acquire irqs and clocks for all physical channels */
+       for (i = 0; i < pdata->num_phy_channels; i++) {
+               struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
+               char clk_name[6];
+
+               phy->id = i;
+               phy->base = s3cdma->base + (i * sdata->stride);
+               phy->host = s3cdma;
+
+               phy->irq = platform_get_irq(pdev, i);
+               if (phy->irq < 0) {
+                       dev_err(&pdev->dev, "failed to get irq %d, err %d\n",
+                               i, phy->irq);
+                       continue;
+               }
+
+               ret = devm_request_irq(&pdev->dev, phy->irq, s3c24xx_dma_irq,
+                                      0, pdev->name, phy);
+               if (ret) {
+                       dev_err(&pdev->dev, "Unable to request irq for channel %d, error %d\n",
+                               i, ret);
+                       continue;
+               }
+
+               if (sdata->has_clocks) {
+                       sprintf(clk_name, "dma.%d", i);
+                       phy->clk = devm_clk_get(&pdev->dev, clk_name);
+                       if (IS_ERR(phy->clk) && sdata->has_clocks) {
+                               dev_err(&pdev->dev, "unable to acquire clock for channel %d, error %lu\n",
+                                       i, PTR_ERR(phy->clk));
+                               continue;
+                       }
+
+                       ret = clk_prepare(phy->clk);
+                       if (ret) {
+                               dev_err(&pdev->dev, "clock for phy %d failed, error %d\n",
+                                       i, ret);
+                               continue;
+                       }
+               }
+
+               spin_lock_init(&phy->lock);
+               phy->valid = true;
+
+               dev_dbg(&pdev->dev, "physical channel %d is %s\n",
+                       i, s3c24xx_dma_phy_busy(phy) ? "BUSY" : "FREE");
+       }
+
+       /* Initialize memcpy engine */
+       dma_cap_set(DMA_MEMCPY, s3cdma->memcpy.cap_mask);
+       dma_cap_set(DMA_PRIVATE, s3cdma->memcpy.cap_mask);
+       s3cdma->memcpy.dev = &pdev->dev;
+       s3cdma->memcpy.device_free_chan_resources =
+                                       s3c24xx_dma_free_chan_resources;
+       s3cdma->memcpy.device_prep_dma_memcpy = s3c24xx_dma_prep_memcpy;
+       s3cdma->memcpy.device_tx_status = s3c24xx_dma_tx_status;
+       s3cdma->memcpy.device_issue_pending = s3c24xx_dma_issue_pending;
+       s3cdma->memcpy.device_config = s3c24xx_dma_set_runtime_config;
+       s3cdma->memcpy.device_terminate_all = s3c24xx_dma_terminate_all;
+
+       /* Initialize slave engine for SoC internal dedicated peripherals */
+       dma_cap_set(DMA_SLAVE, s3cdma->slave.cap_mask);
+       dma_cap_set(DMA_CYCLIC, s3cdma->slave.cap_mask);
+       dma_cap_set(DMA_PRIVATE, s3cdma->slave.cap_mask);
+       s3cdma->slave.dev = &pdev->dev;
+       s3cdma->slave.device_free_chan_resources =
+                                       s3c24xx_dma_free_chan_resources;
+       s3cdma->slave.device_tx_status = s3c24xx_dma_tx_status;
+       s3cdma->slave.device_issue_pending = s3c24xx_dma_issue_pending;
+       s3cdma->slave.device_prep_slave_sg = s3c24xx_dma_prep_slave_sg;
+       s3cdma->slave.device_prep_dma_cyclic = s3c24xx_dma_prep_dma_cyclic;
+       s3cdma->slave.device_config = s3c24xx_dma_set_runtime_config;
+       s3cdma->slave.device_terminate_all = s3c24xx_dma_terminate_all;
+
+       /* Register as many memcpy channels as there are physical channels */
+       ret = s3c24xx_dma_init_virtual_channels(s3cdma, &s3cdma->memcpy,
+                                               pdata->num_phy_channels, false);
+       if (ret <= 0) {
+               dev_warn(&pdev->dev,
+                        "%s failed to enumerate memcpy channels - %d\n",
+                        __func__, ret);
+               goto err_memcpy;
+       }
+
+       /* Register slave channels */
+       ret = s3c24xx_dma_init_virtual_channels(s3cdma, &s3cdma->slave,
+                               pdata->num_channels, true);
+       if (ret <= 0) {
+               dev_warn(&pdev->dev,
+                       "%s failed to enumerate slave channels - %d\n",
+                               __func__, ret);
+               goto err_slave;
+       }
+
+       ret = dma_async_device_register(&s3cdma->memcpy);
+       if (ret) {
+               dev_warn(&pdev->dev,
+                       "%s failed to register memcpy as an async device - %d\n",
+                       __func__, ret);
+               goto err_memcpy_reg;
+       }
+
+       ret = dma_async_device_register(&s3cdma->slave);
+       if (ret) {
+               dev_warn(&pdev->dev,
+                       "%s failed to register slave as an async device - %d\n",
+                       __func__, ret);
+               goto err_slave_reg;
+       }
+
+       platform_set_drvdata(pdev, s3cdma);
+       dev_info(&pdev->dev, "Loaded dma driver with %d physical channels\n",
+                pdata->num_phy_channels);
+
+       return 0;
+
+err_slave_reg:
+       dma_async_device_unregister(&s3cdma->memcpy);
+err_memcpy_reg:
+       s3c24xx_dma_free_virtual_channels(&s3cdma->slave);
+err_slave:
+       s3c24xx_dma_free_virtual_channels(&s3cdma->memcpy);
+err_memcpy:
+       if (sdata->has_clocks)
+               for (i = 0; i < pdata->num_phy_channels; i++) {
+                       struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
+                       if (phy->valid)
+                               clk_unprepare(phy->clk);
+               }
+
+       return ret;
+}
+
+static int s3c24xx_dma_remove(struct platform_device *pdev)
+{
+       const struct s3c24xx_dma_platdata *pdata = dev_get_platdata(&pdev->dev);
+       struct s3c24xx_dma_engine *s3cdma = platform_get_drvdata(pdev);
+       struct soc_data *sdata = s3c24xx_dma_get_soc_data(pdev);
+       int i;
+
+       dma_async_device_unregister(&s3cdma->slave);
+       dma_async_device_unregister(&s3cdma->memcpy);
+
+       s3c24xx_dma_free_virtual_channels(&s3cdma->slave);
+       s3c24xx_dma_free_virtual_channels(&s3cdma->memcpy);
+
+       if (sdata->has_clocks)
+               for (i = 0; i < pdata->num_phy_channels; i++) {
+                       struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
+                       if (phy->valid)
+                               clk_unprepare(phy->clk);
+               }
+
+       return 0;
+}
+
+static struct platform_driver s3c24xx_dma_driver = {
+       .driver         = {
+               .name   = "s3c24xx-dma",
+       },
+       .id_table       = s3c24xx_dma_driver_ids,
+       .probe          = s3c24xx_dma_probe,
+       .remove         = s3c24xx_dma_remove,
+};
+
+module_platform_driver(s3c24xx_dma_driver);
+
+bool s3c24xx_dma_filter(struct dma_chan *chan, void *param)
+{
+       struct s3c24xx_dma_chan *s3cchan;
+
+       if (chan->device->dev->driver != &s3c24xx_dma_driver.driver)
+               return false;
+
+       s3cchan = to_s3c24xx_dma_chan(chan);
+
+       return s3cchan->id == (int)param;
+}
+EXPORT_SYMBOL(s3c24xx_dma_filter);
+
+MODULE_DESCRIPTION("S3C24XX DMA Driver");
+MODULE_AUTHOR("Heiko Stuebner");
+MODULE_LICENSE("GPL v2");