Patchwork [01/14] dmaengine: imx-dma: merge old dma-v1.c with imx-dma.c

login
register
mail settings
Submitter Javier Martin
Date March 5, 2012, 4:16 p.m.
Message ID <1330964196-32623-2-git-send-email-javier.martin@vista-silicon.com>
Download mbox | patch
Permalink /patch/144710/
State New
Headers show

Comments

Javier Martin - March 5, 2012, 4:16 p.m.
Signed-off-by: Javier Martin <javier.martin@vista-silicon.com>
---
 arch/arm/mach-imx/Kconfig  |    6 -
 arch/arm/mach-imx/Makefile |    2 -
 drivers/dma/Kconfig        |    1 -
 drivers/dma/imx-dma.c      |  591 ++++++++++++++++++++++++++++++++++++++++----
 4 files changed, 541 insertions(+), 59 deletions(-)
Koul, Vinod - March 7, 2012, 6:42 a.m.
On Mon, 2012-03-05 at 17:16 +0100, Javier Martin wrote:
Sascha,
Are there any more users of dmac in mach-imx? If not then why not move
the code, rather than copy. If yes can we move them as well?

> Signed-off-by: Javier Martin <javier.martin@vista-silicon.com>
> ---
>  arch/arm/mach-imx/Kconfig  |    6 -
>  arch/arm/mach-imx/Makefile |    2 -
>  drivers/dma/Kconfig        |    1 -
>  drivers/dma/imx-dma.c      |  591 ++++++++++++++++++++++++++++++++++++++++----
>  4 files changed, 541 insertions(+), 59 deletions(-)
> 
> diff --git a/arch/arm/mach-imx/Kconfig b/arch/arm/mach-imx/Kconfig
> index 0e6de36..3da1421 100644
> --- a/arch/arm/mach-imx/Kconfig
> +++ b/arch/arm/mach-imx/Kconfig
> @@ -1,6 +1,3 @@
> -config IMX_HAVE_DMA_V1
> -	bool
> -
>  config HAVE_IMX_GPC
>  	bool
>  
> @@ -26,7 +23,6 @@ config SOC_IMX1
>  	bool
>  	select ARCH_MX1
>  	select CPU_ARM920T
> -	select IMX_HAVE_DMA_V1
>  	select IMX_HAVE_IOMUX_V1
>  	select MXC_AVIC
>  
> @@ -35,7 +31,6 @@ config SOC_IMX21
>  	select MACH_MX21
>  	select CPU_ARM926T
>  	select ARCH_MXC_AUDMUX_V1
> -	select IMX_HAVE_DMA_V1
>  	select IMX_HAVE_IOMUX_V1
>  	select MXC_AVIC
>  
> @@ -52,7 +47,6 @@ config SOC_IMX27
>  	select MACH_MX27
>  	select CPU_ARM926T
>  	select ARCH_MXC_AUDMUX_V1
> -	select IMX_HAVE_DMA_V1
>  	select IMX_HAVE_IOMUX_V1
>  	select MXC_AVIC
>  
> diff --git a/arch/arm/mach-imx/Makefile b/arch/arm/mach-imx/Makefile
> index f5920c2..a62dc3a 100644
> --- a/arch/arm/mach-imx/Makefile
> +++ b/arch/arm/mach-imx/Makefile
> @@ -1,5 +1,3 @@
> -obj-$(CONFIG_IMX_HAVE_DMA_V1) += dma-v1.o
> -
>  obj-$(CONFIG_SOC_IMX1) += clock-imx1.o mm-imx1.o
>  obj-$(CONFIG_SOC_IMX21) += clock-imx21.o mm-imx21.o
>  
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index f1a2749..eb58b91 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -231,7 +231,6 @@ config IMX_SDMA
>  
>  config IMX_DMA
>  	tristate "i.MX DMA support"
> -	depends on IMX_HAVE_DMA_V1
>  	select DMA_ENGINE
>  	help
>  	  Support the i.MX DMA engine. This engine is integrated into
> diff --git a/drivers/dma/imx-dma.c b/drivers/dma/imx-dma.c
> index c32103f..28d716c 100644
> --- a/drivers/dma/imx-dma.c
> +++ b/drivers/dma/imx-dma.c
> @@ -14,7 +14,6 @@
>   * http://www.opensource.org/licenses/gpl-license.html
>   * http://www.gnu.org/copyleft/gpl.html
>   */
> -
>  #include <linux/init.h>
>  #include <linux/module.h>
>  #include <linux/types.h>
> @@ -25,14 +24,88 @@
>  #include <linux/dma-mapping.h>
>  #include <linux/slab.h>
>  #include <linux/platform_device.h>
> +#include <linux/clk.h>
>  #include <linux/dmaengine.h>
>  #include <linux/module.h>
>  
>  #include <asm/irq.h>
> -#include <mach/dma-v1.h>
> +#include <mach/dma.h>
>  #include <mach/hardware.h>
>  
>  #define IMXDMA_MAX_CHAN_DESCRIPTORS	16
> +#define IMX_DMA_CHANNELS  16
> +
> +#define DMA_MODE_READ		0
> +#define DMA_MODE_WRITE		1
> +#define DMA_MODE_MASK		1
> +
> +#define IMX_DMA_LENGTH_LOOP	((unsigned int)-1)
> +#define IMX_DMA_MEMSIZE_32	(0 << 4)
> +#define IMX_DMA_MEMSIZE_8	(1 << 4)
> +#define IMX_DMA_MEMSIZE_16	(2 << 4)
> +#define IMX_DMA_TYPE_LINEAR	(0 << 10)
> +#define IMX_DMA_TYPE_2D		(1 << 10)
> +#define IMX_DMA_TYPE_FIFO	(2 << 10)
> +
> +#define IMX_DMA_ERR_BURST     (1 << 0)
> +#define IMX_DMA_ERR_REQUEST   (1 << 1)
> +#define IMX_DMA_ERR_TRANSFER  (1 << 2)
> +#define IMX_DMA_ERR_BUFFER    (1 << 3)
> +#define IMX_DMA_ERR_TIMEOUT   (1 << 4)
> +
> +#define DMA_DCR     0x00		/* Control Register */
> +#define DMA_DISR    0x04		/* Interrupt status Register */
> +#define DMA_DIMR    0x08		/* Interrupt mask Register */
> +#define DMA_DBTOSR  0x0c		/* Burst timeout status Register */
> +#define DMA_DRTOSR  0x10		/* Request timeout Register */
> +#define DMA_DSESR   0x14		/* Transfer Error Status Register */
> +#define DMA_DBOSR   0x18		/* Buffer overflow status Register */
> +#define DMA_DBTOCR  0x1c		/* Burst timeout control Register */
> +#define DMA_WSRA    0x40		/* W-Size Register A */
> +#define DMA_XSRA    0x44		/* X-Size Register A */
> +#define DMA_YSRA    0x48		/* Y-Size Register A */
> +#define DMA_WSRB    0x4c		/* W-Size Register B */
> +#define DMA_XSRB    0x50		/* X-Size Register B */
> +#define DMA_YSRB    0x54		/* Y-Size Register B */
> +#define DMA_SAR(x)  (0x80 + ((x) << 6))	/* Source Address Registers */
> +#define DMA_DAR(x)  (0x84 + ((x) << 6))	/* Destination Address Registers */
> +#define DMA_CNTR(x) (0x88 + ((x) << 6))	/* Count Registers */
> +#define DMA_CCR(x)  (0x8c + ((x) << 6))	/* Control Registers */
> +#define DMA_RSSR(x) (0x90 + ((x) << 6))	/* Request source select Registers */
> +#define DMA_BLR(x)  (0x94 + ((x) << 6))	/* Burst length Registers */
> +#define DMA_RTOR(x) (0x98 + ((x) << 6))	/* Request timeout Registers */
> +#define DMA_BUCR(x) (0x98 + ((x) << 6))	/* Bus Utilization Registers */
> +#define DMA_CCNR(x) (0x9C + ((x) << 6))	/* Channel counter Registers */
> +
> +#define DCR_DRST           (1<<1)
> +#define DCR_DEN            (1<<0)
> +#define DBTOCR_EN          (1<<15)
> +#define DBTOCR_CNT(x)      ((x) & 0x7fff)
> +#define CNTR_CNT(x)        ((x) & 0xffffff)
> +#define CCR_ACRPT          (1<<14)
> +#define CCR_DMOD_LINEAR    (0x0 << 12)
> +#define CCR_DMOD_2D        (0x1 << 12)
> +#define CCR_DMOD_FIFO      (0x2 << 12)
> +#define CCR_DMOD_EOBFIFO   (0x3 << 12)
> +#define CCR_SMOD_LINEAR    (0x0 << 10)
> +#define CCR_SMOD_2D        (0x1 << 10)
> +#define CCR_SMOD_FIFO      (0x2 << 10)
> +#define CCR_SMOD_EOBFIFO   (0x3 << 10)
> +#define CCR_MDIR_DEC       (1<<9)
> +#define CCR_MSEL_B         (1<<8)
> +#define CCR_DSIZ_32        (0x0 << 6)
> +#define CCR_DSIZ_8         (0x1 << 6)
> +#define CCR_DSIZ_16        (0x2 << 6)
> +#define CCR_SSIZ_32        (0x0 << 4)
> +#define CCR_SSIZ_8         (0x1 << 4)
> +#define CCR_SSIZ_16        (0x2 << 4)
> +#define CCR_REN            (1<<3)
> +#define CCR_RPT            (1<<2)
> +#define CCR_FRC            (1<<1)
> +#define CCR_CEN            (1<<0)
> +#define RTOR_EN            (1<<15)
> +#define RTOR_CLK           (1<<14)
> +#define RTOR_PSC           (1<<13)
>  
>  enum  imxdma_prep_type {
>  	IMXDMA_DESC_MEMCPY,
> @@ -41,6 +114,39 @@ enum  imxdma_prep_type {
>  	IMXDMA_DESC_CYCLIC,
>  };
>  
> +/*
> + * struct imxdma_channel_internal - i.MX specific DMA extension
> + * @name: name specified by DMA client
> + * @irq_handler: client callback for end of transfer
> + * @err_handler: client callback for error condition
> + * @data: clients context data for callbacks
> + * @dma_mode: direction of the transfer %DMA_MODE_READ or %DMA_MODE_WRITE
> + * @sg: pointer to the actual read/written chunk for scatter-gather emulation
> + * @resbytes: total residual number of bytes to transfer
> + *            (it can be lower or same as sum of SG mapped chunk sizes)
> + * @sgcount: number of chunks to be read/written
> + *
> + * Structure is used for IMX DMA processing. It would be probably good
> + * @struct dma_struct in the future for external interfacing and use
> + * @struct imxdma_channel_internal only as extension to it.
> + */
> +
> +struct imxdma_channel_internal {
> +	void *data;
> +	unsigned int dma_mode;
> +	struct scatterlist *sg;
> +	unsigned int resbytes;
> +
> +	int in_use;
> +
> +	u32 ccr_from_device;
> +	u32 ccr_to_device;
> +
> +	struct timer_list watchdog;
> +
> +	int hw_chaining;
> +};
> +
>  struct imxdma_desc {
>  	struct list_head		node;
>  	struct dma_async_tx_descriptor	desc;
> @@ -63,9 +169,9 @@ struct imxdma_desc {
>  };
>  
>  struct imxdma_channel {
> +	struct imxdma_channel_internal	internal;
>  	struct imxdma_engine		*imxdma;
>  	unsigned int			channel;
> -	unsigned int			imxdma_channel;
>  
>  	struct tasklet_struct		dma_tasklet;
>  	struct list_head		ld_free;
> @@ -109,28 +215,381 @@ static inline bool imxdma_chan_is_doing_cyclic(struct imxdma_channel *imxdmac)
>  	return false;
>  }
>  
> -static void imxdma_irq_handler(int channel, void *data)
> +/* TODO: put this inside any struct */
> +static void __iomem *imx_dmav1_baseaddr;
> +static struct clk *dma_clk;
> +
> +static void imx_dmav1_writel(unsigned val, unsigned offset)
>  {
> -	struct imxdma_channel *imxdmac = data;
> +	__raw_writel(val, imx_dmav1_baseaddr + offset);
> +}
>  
> -	tasklet_schedule(&imxdmac->dma_tasklet);
> +static unsigned imx_dmav1_readl(unsigned offset)
> +{
> +	return __raw_readl(imx_dmav1_baseaddr + offset);
>  }
>  
> -static void imxdma_err_handler(int channel, void *data, int error)
> +static int imxdma_hw_chain(struct imxdma_channel_internal *imxdma)
>  {
> -	struct imxdma_channel *imxdmac = data;
> +	if (cpu_is_mx27())
> +		return imxdma->hw_chaining;
> +	else
> +		return 0;
> +}
> +
> +/*
> + * imxdma_sg_next - prepare next chunk for scatter-gather DMA emulation
> + */
> +static inline int imxdma_sg_next(struct imxdma_channel *imxdmac, struct scatterlist *sg)
> +{
> +	struct imxdma_channel_internal *imxdma = &imxdmac->internal;
> +	unsigned long now;
> +
> +	now = min(imxdma->resbytes, sg->length);
> +	if (imxdma->resbytes != IMX_DMA_LENGTH_LOOP)
> +		imxdma->resbytes -= now;
> +
> +	if ((imxdma->dma_mode & DMA_MODE_MASK) == DMA_MODE_READ)
> +		imx_dmav1_writel(sg->dma_address, DMA_DAR(imxdmac->channel));
> +	else
> +		imx_dmav1_writel(sg->dma_address, DMA_SAR(imxdmac->channel));
> +
> +	imx_dmav1_writel(now, DMA_CNTR(imxdmac->channel));
> +
> +	pr_debug("imxdma%d: next sg chunk dst 0x%08x, src 0x%08x, "
> +		"size 0x%08x\n", imxdmac->channel,
> +		 imx_dmav1_readl(DMA_DAR(imxdmac->channel)),
> +		 imx_dmav1_readl(DMA_SAR(imxdmac->channel)),
> +		 imx_dmav1_readl(DMA_CNTR(imxdmac->channel)));
> +
> +	return now;
> +}
> +
> +static int
> +imxdma_setup_single_hw(struct imxdma_channel *imxdmac, dma_addr_t dma_address,
> +		     unsigned int dma_length, unsigned int dev_addr,
> +		     unsigned int dmamode)
> +{
> +	int channel = imxdmac->channel;
> +
> +	imxdmac->internal.sg = NULL;
> +	imxdmac->internal.dma_mode = dmamode;
> +
> +	if (!dma_address) {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_single null address\n",
> +		       channel);
> +		return -EINVAL;
> +	}
> +
> +	if (!dma_length) {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_single zero length\n",
> +		       channel);
> +		return -EINVAL;
> +	}
> +
> +	if ((dmamode & DMA_MODE_MASK) == DMA_MODE_READ) {
> +		pr_debug("imxdma%d: %s dma_addressg=0x%08x dma_length=%d "
> +			"dev_addr=0x%08x for read\n",
> +			channel, __func__, (unsigned int)dma_address,
> +			dma_length, dev_addr);
> +
> +		imx_dmav1_writel(dev_addr, DMA_SAR(channel));
> +		imx_dmav1_writel(dma_address, DMA_DAR(channel));
> +		imx_dmav1_writel(imxdmac->internal.ccr_from_device, DMA_CCR(channel));
> +	} else if ((dmamode & DMA_MODE_MASK) == DMA_MODE_WRITE) {
> +		pr_debug("imxdma%d: %s dma_addressg=0x%08x dma_length=%d "
> +			"dev_addr=0x%08x for write\n",
> +			channel, __func__, (unsigned int)dma_address,
> +			dma_length, dev_addr);
> +
> +		imx_dmav1_writel(dma_address, DMA_SAR(channel));
> +		imx_dmav1_writel(dev_addr, DMA_DAR(channel));
> +		imx_dmav1_writel(imxdmac->internal.ccr_to_device,
> +				DMA_CCR(channel));
> +	} else {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_single bad dmamode\n",
> +		       channel);
> +		return -EINVAL;
> +	}
>  
> +	imx_dmav1_writel(dma_length, DMA_CNTR(channel));
> +
> +	return 0;
> +}
> +
> +static void imxdma_enable_hw(struct imxdma_channel *imxdmac)
> +{
> +	int channel = imxdmac->channel;
> +	unsigned long flags;
> +
> +	pr_debug("imxdma%d: imx_dma_enable\n", channel);
> +
> +	if (imxdmac->internal.in_use)
> +		return;
> +
> +	local_irq_save(flags);
> +
> +	imx_dmav1_writel(1 << channel, DMA_DISR);
> +	imx_dmav1_writel(imx_dmav1_readl(DMA_DIMR) & ~(1 << channel), DMA_DIMR);
> +	imx_dmav1_writel(imx_dmav1_readl(DMA_CCR(channel)) | CCR_CEN |
> +		CCR_ACRPT, DMA_CCR(channel));
> +
> +	if ((cpu_is_mx21() || cpu_is_mx27()) &&
> +			imxdmac->internal.sg && imxdma_hw_chain(&imxdmac->internal)) {
> +		imxdmac->internal.sg = sg_next(imxdmac->internal.sg);
> +		if (imxdmac->internal.sg) {
> +			u32 tmp;
> +			imxdma_sg_next(imxdmac, imxdmac->internal.sg);
> +			tmp = imx_dmav1_readl(DMA_CCR(channel));
> +			imx_dmav1_writel(tmp | CCR_RPT | CCR_ACRPT,
> +				DMA_CCR(channel));
> +		}
> +	}
> +	imxdmac->internal.in_use = 1;
> +
> +	local_irq_restore(flags);
> +}
> +
> +static void imxdma_disable_hw(struct imxdma_channel *imxdmac)
> +{
> +	int channel = imxdmac->channel;
> +	unsigned long flags;
> +
> +	pr_debug("imxdma%d: imx_dma_disable\n", channel);
> +
> +	if (imxdma_hw_chain(&imxdmac->internal))
> +		del_timer(&imxdmac->internal.watchdog);
> +
> +	local_irq_save(flags);
> +	imx_dmav1_writel(imx_dmav1_readl(DMA_DIMR) | (1 << channel), DMA_DIMR);
> +	imx_dmav1_writel(imx_dmav1_readl(DMA_CCR(channel)) & ~CCR_CEN,
> +			DMA_CCR(channel));
> +	imx_dmav1_writel(1 << channel, DMA_DISR);
> +	imxdmac->internal.in_use = 0;
> +	local_irq_restore(flags);
> +}
> +
> +static int
> +imxdma_config_channel_hw(struct imxdma_channel *imxdmac, unsigned int config_port,
> +	unsigned int config_mem, unsigned int dmareq, int hw_chaining)
> +{
> +	int channel = imxdmac->channel;
> +	u32 dreq = 0;
> +
> +	imxdmac->internal.hw_chaining = 0;
> +
> +	if (hw_chaining) {
> +		imxdmac->internal.hw_chaining = 1;
> +		if (!imxdma_hw_chain(&imxdmac->internal))
> +			return -EINVAL;
> +	}
> +
> +	if (dmareq)
> +		dreq = CCR_REN;
> +
> +	imxdmac->internal.ccr_from_device = config_port | (config_mem << 2) | dreq;
> +	imxdmac->internal.ccr_to_device = config_mem | (config_port << 2) | dreq;
> +
> +	imx_dmav1_writel(dmareq, DMA_RSSR(channel));
> +
> +	return 0;
> +}
> +
> +static int
> +imxdma_setup_sg_hw(struct imxdma_channel *imxdmac,
> +		 struct scatterlist *sg, unsigned int sgcount,
> +		 unsigned int dma_length, unsigned int dev_addr,
> +		 unsigned int dmamode)
> +{
> +	int channel = imxdmac->channel;
> +
> +	if (imxdmac->internal.in_use)
> +		return -EBUSY;
> +
> +	imxdmac->internal.sg = sg;
> +	imxdmac->internal.dma_mode = dmamode;
> +	imxdmac->internal.resbytes = dma_length;
> +
> +	if (!sg || !sgcount) {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg empty sg list\n",
> +		       channel);
> +		return -EINVAL;
> +	}
> +
> +	if (!sg->length) {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg zero length\n",
> +		       channel);
> +		return -EINVAL;
> +	}
> +
> +	if ((dmamode & DMA_MODE_MASK) == DMA_MODE_READ) {
> +		pr_debug("imxdma%d: %s sg=%p sgcount=%d total length=%d "
> +			"dev_addr=0x%08x for read\n",
> +			channel, __func__, sg, sgcount, dma_length, dev_addr);
> +
> +		imx_dmav1_writel(dev_addr, DMA_SAR(channel));
> +		imx_dmav1_writel(imxdmac->internal.ccr_from_device, DMA_CCR(channel));
> +	} else if ((dmamode & DMA_MODE_MASK) == DMA_MODE_WRITE) {
> +		pr_debug("imxdma%d: %s sg=%p sgcount=%d total length=%d "
> +			"dev_addr=0x%08x for write\n",
> +			channel, __func__, sg, sgcount, dma_length, dev_addr);
> +
> +		imx_dmav1_writel(dev_addr, DMA_DAR(channel));
> +		imx_dmav1_writel(imxdmac->internal.ccr_to_device, DMA_CCR(channel));
> +	} else {
> +		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg bad dmamode\n",
> +		       channel);
> +		return -EINVAL;
> +	}
> +
> +	imxdma_sg_next(imxdmac, sg);
> +
> +	return 0;
> +}
> +
> +static void imxdma_watchdog(unsigned long data)
> +{
> +	struct imxdma_channel *imxdmac = (struct imxdma_channel *)data;
> +	int channel = imxdmac->channel;
> +
> +	imx_dmav1_writel(0, DMA_CCR(channel));
> +	imxdmac->internal.in_use = 0;
> +	imxdmac->internal.sg = NULL;
> +
> +	/* Tasklet watchdog error handler */
>  	tasklet_schedule(&imxdmac->dma_tasklet);
> +	pr_debug("imxdma%d: watchdog timeout!\n", imxdmac->channel);
> +}
> +
> +static irqreturn_t imxdma_err_handler(int irq, void *dev_id)
> +{
> +	struct imxdma_engine *imxdma = dev_id;
> +	struct imxdma_channel_internal *internal;
> +	unsigned int err_mask;
> +	int i, disr;
> +	int errcode;
> +
> +	disr = imx_dmav1_readl(DMA_DISR);
> +
> +	err_mask = imx_dmav1_readl(DMA_DBTOSR) |
> +		   imx_dmav1_readl(DMA_DRTOSR) |
> +		   imx_dmav1_readl(DMA_DSESR)  |
> +		   imx_dmav1_readl(DMA_DBOSR);
> +
> +	if (!err_mask)
> +		return IRQ_HANDLED;
> +
> +	imx_dmav1_writel(disr & err_mask, DMA_DISR);
> +
> +	for (i = 0; i < IMX_DMA_CHANNELS; i++) {
> +		if (!(err_mask & (1 << i)))
> +			continue;
> +		internal = &imxdma->channel[i].internal;
> +		errcode = 0;
> +
> +		if (imx_dmav1_readl(DMA_DBTOSR) & (1 << i)) {
> +			imx_dmav1_writel(1 << i, DMA_DBTOSR);
> +			errcode |= IMX_DMA_ERR_BURST;
> +		}
> +		if (imx_dmav1_readl(DMA_DRTOSR) & (1 << i)) {
> +			imx_dmav1_writel(1 << i, DMA_DRTOSR);
> +			errcode |= IMX_DMA_ERR_REQUEST;
> +		}
> +		if (imx_dmav1_readl(DMA_DSESR) & (1 << i)) {
> +			imx_dmav1_writel(1 << i, DMA_DSESR);
> +			errcode |= IMX_DMA_ERR_TRANSFER;
> +		}
> +		if (imx_dmav1_readl(DMA_DBOSR) & (1 << i)) {
> +			imx_dmav1_writel(1 << i, DMA_DBOSR);
> +			errcode |= IMX_DMA_ERR_BUFFER;
> +		}
> +		/* Tasklet error handler */
> +		tasklet_schedule(&imxdma->channel[i].dma_tasklet);
> +
> +		printk(KERN_WARNING
> +		       "DMA timeout on channel %d -%s%s%s%s\n", i,
> +		       errcode & IMX_DMA_ERR_BURST ?    " burst" : "",
> +		       errcode & IMX_DMA_ERR_REQUEST ?  " request" : "",
> +		       errcode & IMX_DMA_ERR_TRANSFER ? " transfer" : "",
> +		       errcode & IMX_DMA_ERR_BUFFER ?   " buffer" : "");
> +	}
> +	return IRQ_HANDLED;
>  }
>  
> -static void imxdma_progression(int channel, void *data,
> -		struct scatterlist *sg)
> +static void dma_irq_handle_channel(struct imxdma_channel *imxdmac)
>  {
> -	struct imxdma_channel *imxdmac = data;
> +	struct imxdma_channel_internal *imxdma = &imxdmac->internal;
> +	int chno = imxdmac->channel;
> +
> +	if (imxdma->sg) {
> +		u32 tmp;
> +		imxdma->sg = sg_next(imxdma->sg);
> +
> +		if (imxdma->sg) {
> +			imxdma_sg_next(imxdmac, imxdma->sg);
> +
> +			tmp = imx_dmav1_readl(DMA_CCR(chno));
> +
> +			if (imxdma_hw_chain(imxdma)) {
> +				/* FIXME: The timeout should probably be
> +				 * configurable
> +				 */
> +				mod_timer(&imxdma->watchdog,
> +					jiffies + msecs_to_jiffies(500));
> +
> +				tmp |= CCR_CEN | CCR_RPT | CCR_ACRPT;
> +				imx_dmav1_writel(tmp, DMA_CCR(chno));
> +			} else {
> +				imx_dmav1_writel(tmp & ~CCR_CEN, DMA_CCR(chno));
> +				tmp |= CCR_CEN;
> +			}
> +
> +			imx_dmav1_writel(tmp, DMA_CCR(chno));
> +
> +			if (imxdma_chan_is_doing_cyclic(imxdmac))
> +				/* Tasklet progression */
> +				tasklet_schedule(&imxdmac->dma_tasklet);
> +
> +			return;
> +		}
>  
> +		if (imxdma_hw_chain(imxdma)) {
> +			del_timer(&imxdma->watchdog);
> +			return;
> +		}
> +	}
> +
> +	imx_dmav1_writel(0, DMA_CCR(chno));
> +	imxdma->in_use = 0;
> +	/* Tasklet irq */
>  	tasklet_schedule(&imxdmac->dma_tasklet);
>  }
>  
> +static irqreturn_t dma_irq_handler(int irq, void *dev_id)
> +{
> +	struct imxdma_engine *imxdma = dev_id;
> +	struct imxdma_channel_internal *internal;
> +	int i, disr;
> +
> +	if (cpu_is_mx21() || cpu_is_mx27())
> +		imxdma_err_handler(irq, dev_id);
> +
> +	disr = imx_dmav1_readl(DMA_DISR);
> +
> +	pr_debug("imxdma: dma_irq_handler called, disr=0x%08x\n",
> +		     disr);
> +
> +	imx_dmav1_writel(disr, DMA_DISR);
> +	for (i = 0; i < IMX_DMA_CHANNELS; i++) {
> +		if (disr & (1 << i)) {
> +			internal = &imxdma->channel[i].internal;
> +			dma_irq_handle_channel(&imxdma->channel[i]);
> +		}
> +	}
> +
> +	return IRQ_HANDLED;
> +}
> +
>  static int imxdma_xfer_desc(struct imxdma_desc *d)
>  {
>  	struct imxdma_channel *imxdmac = to_imxdma_chan(d->desc.chan);
> @@ -139,31 +598,24 @@ static int imxdma_xfer_desc(struct imxdma_desc *d)
>  	/* Configure and enable */
>  	switch (d->type) {
>  	case IMXDMA_DESC_MEMCPY:
> -		ret = imx_dma_config_channel(imxdmac->imxdma_channel,
> +		ret = imxdma_config_channel_hw(imxdmac,
>  					  d->config_port, d->config_mem, 0, 0);
>  		if (ret < 0)
>  			return ret;
> -		ret = imx_dma_setup_single(imxdmac->imxdma_channel, d->src,
> +		ret = imxdma_setup_single_hw(imxdmac, d->src,
>  					   d->len, d->dest, d->dmamode);
>  		if (ret < 0)
>  			return ret;
>  		break;
> +
> +	/* Cyclic transfer is the same as slave_sg with special sg configuration. */
>  	case IMXDMA_DESC_CYCLIC:
> -		ret = imx_dma_setup_progression_handler(imxdmac->imxdma_channel,
> -							imxdma_progression);
> -		if (ret < 0)
> -			return ret;
> -		/*
> -		 * We fall through here since cyclic transfer is the same as
> -		 * slave_sg adding a progression handler and a specific sg
> -		 * configuration which is done in 'imxdma_prep_dma_cyclic'.
> -		 */
>  	case IMXDMA_DESC_SLAVE_SG:
>  		if (d->dmamode == DMA_MODE_READ)
> -			ret = imx_dma_setup_sg(imxdmac->imxdma_channel, d->sg,
> +			ret = imxdma_setup_sg_hw(imxdmac, d->sg,
>  				       d->sgcount, d->len, d->src, d->dmamode);
>  		else
> -			ret = imx_dma_setup_sg(imxdmac->imxdma_channel, d->sg,
> +			ret = imxdma_setup_sg_hw(imxdmac, d->sg,
>  				      d->sgcount, d->len, d->dest, d->dmamode);
>  		if (ret < 0)
>  			return ret;
> @@ -171,7 +623,7 @@ static int imxdma_xfer_desc(struct imxdma_desc *d)
>  	default:
>  		return -EINVAL;
>  	}
> -	imx_dma_enable(imxdmac->imxdma_channel);
> +	imxdma_enable_hw(imxdmac);
>  	return 0;
>  }
>  
> @@ -223,7 +675,7 @@ static int imxdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
>  
>  	switch (cmd) {
>  	case DMA_TERMINATE_ALL:
> -		imx_dma_disable(imxdmac->imxdma_channel);
> +		imxdma_disable_hw(imxdmac);
>  
>  		spin_lock_irqsave(&imxdmac->lock, flags);
>  		list_splice_tail_init(&imxdmac->ld_active, &imxdmac->ld_free);
> @@ -253,16 +705,16 @@ static int imxdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
>  			mode = IMX_DMA_MEMSIZE_32;
>  			break;
>  		}
> -		ret = imx_dma_config_channel(imxdmac->imxdma_channel,
> +		ret = imxdma_config_channel_hw(imxdmac,
>  				mode | IMX_DMA_TYPE_FIFO,
>  				IMX_DMA_MEMSIZE_32 | IMX_DMA_TYPE_LINEAR,
>  				imxdmac->dma_request, 1);
>  
>  		if (ret)
>  			return ret;
> -
> -		imx_dma_config_burstlen(imxdmac->imxdma_channel,
> -				imxdmac->watermark_level * imxdmac->word_size);
> +		/* Set burst length */
> +		imx_dmav1_writel(imxdmac->watermark_level * imxdmac->word_size,
> +				 DMA_BLR(imxdmac->channel));
>  
>  		return 0;
>  	default:
> @@ -359,7 +811,7 @@ static void imxdma_free_chan_resources(struct dma_chan *chan)
>  
>  	spin_lock_irqsave(&imxdmac->lock, flags);
>  
> -	imx_dma_disable(imxdmac->imxdma_channel);
> +	imxdma_disable_hw(imxdmac);
>  	list_splice_tail_init(&imxdmac->ld_active, &imxdmac->ld_free);
>  	list_splice_tail_init(&imxdmac->ld_queue, &imxdmac->ld_free);
>  
> @@ -545,10 +997,51 @@ static void imxdma_issue_pending(struct dma_chan *chan)
>  }
>  
>  static int __init imxdma_probe(struct platform_device *pdev)
> -{
> +	{
>  	struct imxdma_engine *imxdma;
>  	int ret, i;
>  
> +	if (cpu_is_mx1())
> +		imx_dmav1_baseaddr = MX1_IO_ADDRESS(MX1_DMA_BASE_ADDR);
> +	else if (cpu_is_mx21())
> +		imx_dmav1_baseaddr = MX21_IO_ADDRESS(MX21_DMA_BASE_ADDR);
> +	else if (cpu_is_mx27())
> +		imx_dmav1_baseaddr = MX27_IO_ADDRESS(MX27_DMA_BASE_ADDR);
> +	else
> +		return 0;
> +
> +	dma_clk = clk_get(NULL, "dma");
> +	if (IS_ERR(dma_clk))
> +		return PTR_ERR(dma_clk);
> +	clk_enable(dma_clk);
> +
> +	/* reset DMA module */
> +	imx_dmav1_writel(DCR_DRST, DMA_DCR);
> +
> +	if (cpu_is_mx1()) {
> +		ret = request_irq(MX1_DMA_INT, dma_irq_handler, 0, "DMA", imxdma);
> +		if (ret) {
> +			pr_crit("Can't register IRQ for DMA\n");
> +			return ret;
> +		}
> +
> +		ret = request_irq(MX1_DMA_ERR, imxdma_err_handler, 0, "DMA", imxdma);
> +		if (ret) {
> +			pr_crit("Can't register ERRIRQ for DMA\n");
> +			free_irq(MX1_DMA_INT, NULL);
> +			return ret;
> +		}
> +	}
> +
> +	/* enable DMA module */
> +	imx_dmav1_writel(DCR_DEN, DMA_DCR);
> +
> +	/* clear all interrupts */
> +	imx_dmav1_writel((1 << IMX_DMA_CHANNELS) - 1, DMA_DISR);
> +
> +	/* disable interrupts */
> +	imx_dmav1_writel((1 << IMX_DMA_CHANNELS) - 1, DMA_DIMR);
> +
>  	imxdma = kzalloc(sizeof(*imxdma), GFP_KERNEL);
>  	if (!imxdma)
>  		return -ENOMEM;
> @@ -562,17 +1055,20 @@ static int __init imxdma_probe(struct platform_device *pdev)
>  	/* Initialize channel parameters */
>  	for (i = 0; i < MAX_DMA_CHANNELS; i++) {
>  		struct imxdma_channel *imxdmac = &imxdma->channel[i];
> -
> -		imxdmac->imxdma_channel = imx_dma_request_by_prio("dmaengine",
> -				DMA_PRIO_MEDIUM);
> -		if ((int)imxdmac->channel < 0) {
> -			ret = -ENODEV;
> -			goto err_init;
> +		memset(&imxdmac->internal, 0, sizeof(imxdmac->internal));
> +		if (cpu_is_mx21() || cpu_is_mx27()) {
> +			ret = request_irq(MX2x_INT_DMACH0 + i,
> +					dma_irq_handler, 0, "DMA", imxdma);
> +			if (ret) {
> +				pr_crit("Can't register IRQ %d for DMA channel %d\n",
> +						MX2x_INT_DMACH0 + i, i);
> +				goto err_init;
> +			}
> +			init_timer(&imxdmac->internal.watchdog);
> +			imxdmac->internal.watchdog.function = &imxdma_watchdog;
> +			imxdmac->internal.watchdog.data = (unsigned long)imxdmac;
>  		}
>  
> -		imx_dma_setup_handlers(imxdmac->imxdma_channel,
> -		       imxdma_irq_handler, imxdma_err_handler, imxdmac);
> -
>  		imxdmac->imxdma = imxdma;
>  		spin_lock_init(&imxdmac->lock);
>  
> @@ -617,10 +1113,8 @@ static int __init imxdma_probe(struct platform_device *pdev)
>  	return 0;
>  
>  err_init:
> -	while (--i >= 0) {
> -		struct imxdma_channel *imxdmac = &imxdma->channel[i];
> -		imx_dma_free(imxdmac->imxdma_channel);
> -	}
> +	while (--i >= 0)
> +		free_irq(MX2x_INT_DMACH0 + i, NULL);
>  
>  	kfree(imxdma);
>  	return ret;
> @@ -633,11 +1127,8 @@ static int __exit imxdma_remove(struct platform_device *pdev)
>  
>          dma_async_device_unregister(&imxdma->dma_device);
>  
> -	for (i = 0; i < MAX_DMA_CHANNELS; i++) {
> -		struct imxdma_channel *imxdmac = &imxdma->channel[i];
> -
> -		 imx_dma_free(imxdmac->imxdma_channel);
> -	}
> +	for (i = 0; i < MAX_DMA_CHANNELS; i++)
> +		free_irq(MX2x_INT_DMACH0 + i, NULL);
>  
>          kfree(imxdma);
>
Javier Martin - March 7, 2012, 7:52 a.m.
Hi Vinod,

On 7 March 2012 07:42, Vinod Koul <vinod.koul@intel.com> wrote:
> On Mon, 2012-03-05 at 17:16 +0100, Javier Martin wrote:
> Sascha,
> Are there any more users of dmac in mach-imx?

No, there are no other users of mach-imx. In fact, it was Sascha's
idea to make this merge [1].

> If not then why not move the code, rather than copy.
> If yes can we move them as well?

I do remove the code in mach-imx in PATCH 14/14. I considered it was
better doing this in the last patch to keep the old code available
during the whole development process.

[1] https://lkml.org/lkml/2012/2/9/171
Koul, Vinod - March 7, 2012, 8:15 a.m.
On Wed, 2012-03-07 at 08:52 +0100, javier Martin wrote:
> Hi Vinod,
> 
> On 7 March 2012 07:42, Vinod Koul <vinod.koul@intel.com> wrote:
> > On Mon, 2012-03-05 at 17:16 +0100, Javier Martin wrote:
> > Sascha,
> > Are there any more users of dmac in mach-imx?
> 
> No, there are no other users of mach-imx. In fact, it was Sascha's
> idea to make this merge [1].
Then please move the file in 1st patch, also it would be easily clear
what parts are added, deleted and moved.

Otherwise the series looks fine except that it needs to have a better
changelog on why you are removing certain field/struct. It helps to
document the motivation of doing certain change, you will find that you
wont remember that six months down the line :)

Also I am looking for Tested-By from people who have use imx-dma.

> 
> > If not then why not move the code, rather than copy.
> > If yes can we move them as well?
> 
> I do remove the code in mach-imx in PATCH 14/14. I considered it was
> better doing this in the last patch to keep the old code available
> during the whole development process.
> 
> [1] https://lkml.org/lkml/2012/2/9/171
>
Javier Martin - March 7, 2012, 9:25 a.m.
On 7 March 2012 09:15, Vinod Koul <vinod.koul@intel.com> wrote:
> On Wed, 2012-03-07 at 08:52 +0100, javier Martin wrote:
>> Hi Vinod,
>>
>> On 7 March 2012 07:42, Vinod Koul <vinod.koul@intel.com> wrote:
>> > On Mon, 2012-03-05 at 17:16 +0100, Javier Martin wrote:
>> > Sascha,
>> > Are there any more users of dmac in mach-imx?
>>
>> No, there are no other users of mach-imx. In fact, it was Sascha's
>> idea to make this merge [1].
> Then please move the file in 1st patch, also it would be easily clear
> what parts are added, deleted and moved.

All right, I will, but I'll wait for Sascha and others to comment on
the series too.

> Otherwise the series looks fine except that it needs to have a better
> changelog on why you are removing certain field/struct. It helps to
> document the motivation of doing certain change, you will find that you
> wont remember that six months down the line :)

Thanks for the suggestion, I'll rewrite my changelogs.

> Also I am looking for Tested-By from people who have use imx-dma.

That would be awesome. I've tested this series with a test battery which covers:
- mxcmmc
- audio
- dmatest module

Patch

diff --git a/arch/arm/mach-imx/Kconfig b/arch/arm/mach-imx/Kconfig
index 0e6de36..3da1421 100644
--- a/arch/arm/mach-imx/Kconfig
+++ b/arch/arm/mach-imx/Kconfig
@@ -1,6 +1,3 @@ 
-config IMX_HAVE_DMA_V1
-	bool
-
 config HAVE_IMX_GPC
 	bool
 
@@ -26,7 +23,6 @@  config SOC_IMX1
 	bool
 	select ARCH_MX1
 	select CPU_ARM920T
-	select IMX_HAVE_DMA_V1
 	select IMX_HAVE_IOMUX_V1
 	select MXC_AVIC
 
@@ -35,7 +31,6 @@  config SOC_IMX21
 	select MACH_MX21
 	select CPU_ARM926T
 	select ARCH_MXC_AUDMUX_V1
-	select IMX_HAVE_DMA_V1
 	select IMX_HAVE_IOMUX_V1
 	select MXC_AVIC
 
@@ -52,7 +47,6 @@  config SOC_IMX27
 	select MACH_MX27
 	select CPU_ARM926T
 	select ARCH_MXC_AUDMUX_V1
-	select IMX_HAVE_DMA_V1
 	select IMX_HAVE_IOMUX_V1
 	select MXC_AVIC
 
diff --git a/arch/arm/mach-imx/Makefile b/arch/arm/mach-imx/Makefile
index f5920c2..a62dc3a 100644
--- a/arch/arm/mach-imx/Makefile
+++ b/arch/arm/mach-imx/Makefile
@@ -1,5 +1,3 @@ 
-obj-$(CONFIG_IMX_HAVE_DMA_V1) += dma-v1.o
-
 obj-$(CONFIG_SOC_IMX1) += clock-imx1.o mm-imx1.o
 obj-$(CONFIG_SOC_IMX21) += clock-imx21.o mm-imx21.o
 
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index f1a2749..eb58b91 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -231,7 +231,6 @@  config IMX_SDMA
 
 config IMX_DMA
 	tristate "i.MX DMA support"
-	depends on IMX_HAVE_DMA_V1
 	select DMA_ENGINE
 	help
 	  Support the i.MX DMA engine. This engine is integrated into
diff --git a/drivers/dma/imx-dma.c b/drivers/dma/imx-dma.c
index c32103f..28d716c 100644
--- a/drivers/dma/imx-dma.c
+++ b/drivers/dma/imx-dma.c
@@ -14,7 +14,6 @@ 
  * http://www.opensource.org/licenses/gpl-license.html
  * http://www.gnu.org/copyleft/gpl.html
  */
-
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/types.h>
@@ -25,14 +24,88 @@ 
 #include <linux/dma-mapping.h>
 #include <linux/slab.h>
 #include <linux/platform_device.h>
+#include <linux/clk.h>
 #include <linux/dmaengine.h>
 #include <linux/module.h>
 
 #include <asm/irq.h>
-#include <mach/dma-v1.h>
+#include <mach/dma.h>
 #include <mach/hardware.h>
 
 #define IMXDMA_MAX_CHAN_DESCRIPTORS	16
+#define IMX_DMA_CHANNELS  16
+
+#define DMA_MODE_READ		0
+#define DMA_MODE_WRITE		1
+#define DMA_MODE_MASK		1
+
+#define IMX_DMA_LENGTH_LOOP	((unsigned int)-1)
+#define IMX_DMA_MEMSIZE_32	(0 << 4)
+#define IMX_DMA_MEMSIZE_8	(1 << 4)
+#define IMX_DMA_MEMSIZE_16	(2 << 4)
+#define IMX_DMA_TYPE_LINEAR	(0 << 10)
+#define IMX_DMA_TYPE_2D		(1 << 10)
+#define IMX_DMA_TYPE_FIFO	(2 << 10)
+
+#define IMX_DMA_ERR_BURST     (1 << 0)
+#define IMX_DMA_ERR_REQUEST   (1 << 1)
+#define IMX_DMA_ERR_TRANSFER  (1 << 2)
+#define IMX_DMA_ERR_BUFFER    (1 << 3)
+#define IMX_DMA_ERR_TIMEOUT   (1 << 4)
+
+#define DMA_DCR     0x00		/* Control Register */
+#define DMA_DISR    0x04		/* Interrupt status Register */
+#define DMA_DIMR    0x08		/* Interrupt mask Register */
+#define DMA_DBTOSR  0x0c		/* Burst timeout status Register */
+#define DMA_DRTOSR  0x10		/* Request timeout Register */
+#define DMA_DSESR   0x14		/* Transfer Error Status Register */
+#define DMA_DBOSR   0x18		/* Buffer overflow status Register */
+#define DMA_DBTOCR  0x1c		/* Burst timeout control Register */
+#define DMA_WSRA    0x40		/* W-Size Register A */
+#define DMA_XSRA    0x44		/* X-Size Register A */
+#define DMA_YSRA    0x48		/* Y-Size Register A */
+#define DMA_WSRB    0x4c		/* W-Size Register B */
+#define DMA_XSRB    0x50		/* X-Size Register B */
+#define DMA_YSRB    0x54		/* Y-Size Register B */
+#define DMA_SAR(x)  (0x80 + ((x) << 6))	/* Source Address Registers */
+#define DMA_DAR(x)  (0x84 + ((x) << 6))	/* Destination Address Registers */
+#define DMA_CNTR(x) (0x88 + ((x) << 6))	/* Count Registers */
+#define DMA_CCR(x)  (0x8c + ((x) << 6))	/* Control Registers */
+#define DMA_RSSR(x) (0x90 + ((x) << 6))	/* Request source select Registers */
+#define DMA_BLR(x)  (0x94 + ((x) << 6))	/* Burst length Registers */
+#define DMA_RTOR(x) (0x98 + ((x) << 6))	/* Request timeout Registers */
+#define DMA_BUCR(x) (0x98 + ((x) << 6))	/* Bus Utilization Registers */
+#define DMA_CCNR(x) (0x9C + ((x) << 6))	/* Channel counter Registers */
+
+#define DCR_DRST           (1<<1)
+#define DCR_DEN            (1<<0)
+#define DBTOCR_EN          (1<<15)
+#define DBTOCR_CNT(x)      ((x) & 0x7fff)
+#define CNTR_CNT(x)        ((x) & 0xffffff)
+#define CCR_ACRPT          (1<<14)
+#define CCR_DMOD_LINEAR    (0x0 << 12)
+#define CCR_DMOD_2D        (0x1 << 12)
+#define CCR_DMOD_FIFO      (0x2 << 12)
+#define CCR_DMOD_EOBFIFO   (0x3 << 12)
+#define CCR_SMOD_LINEAR    (0x0 << 10)
+#define CCR_SMOD_2D        (0x1 << 10)
+#define CCR_SMOD_FIFO      (0x2 << 10)
+#define CCR_SMOD_EOBFIFO   (0x3 << 10)
+#define CCR_MDIR_DEC       (1<<9)
+#define CCR_MSEL_B         (1<<8)
+#define CCR_DSIZ_32        (0x0 << 6)
+#define CCR_DSIZ_8         (0x1 << 6)
+#define CCR_DSIZ_16        (0x2 << 6)
+#define CCR_SSIZ_32        (0x0 << 4)
+#define CCR_SSIZ_8         (0x1 << 4)
+#define CCR_SSIZ_16        (0x2 << 4)
+#define CCR_REN            (1<<3)
+#define CCR_RPT            (1<<2)
+#define CCR_FRC            (1<<1)
+#define CCR_CEN            (1<<0)
+#define RTOR_EN            (1<<15)
+#define RTOR_CLK           (1<<14)
+#define RTOR_PSC           (1<<13)
 
 enum  imxdma_prep_type {
 	IMXDMA_DESC_MEMCPY,
@@ -41,6 +114,39 @@  enum  imxdma_prep_type {
 	IMXDMA_DESC_CYCLIC,
 };
 
+/*
+ * struct imxdma_channel_internal - i.MX specific DMA extension
+ * @name: name specified by DMA client
+ * @irq_handler: client callback for end of transfer
+ * @err_handler: client callback for error condition
+ * @data: clients context data for callbacks
+ * @dma_mode: direction of the transfer %DMA_MODE_READ or %DMA_MODE_WRITE
+ * @sg: pointer to the actual read/written chunk for scatter-gather emulation
+ * @resbytes: total residual number of bytes to transfer
+ *            (it can be lower or same as sum of SG mapped chunk sizes)
+ * @sgcount: number of chunks to be read/written
+ *
+ * Structure is used for IMX DMA processing. It would be probably good
+ * @struct dma_struct in the future for external interfacing and use
+ * @struct imxdma_channel_internal only as extension to it.
+ */
+
+struct imxdma_channel_internal {
+	void *data;
+	unsigned int dma_mode;
+	struct scatterlist *sg;
+	unsigned int resbytes;
+
+	int in_use;
+
+	u32 ccr_from_device;
+	u32 ccr_to_device;
+
+	struct timer_list watchdog;
+
+	int hw_chaining;
+};
+
 struct imxdma_desc {
 	struct list_head		node;
 	struct dma_async_tx_descriptor	desc;
@@ -63,9 +169,9 @@  struct imxdma_desc {
 };
 
 struct imxdma_channel {
+	struct imxdma_channel_internal	internal;
 	struct imxdma_engine		*imxdma;
 	unsigned int			channel;
-	unsigned int			imxdma_channel;
 
 	struct tasklet_struct		dma_tasklet;
 	struct list_head		ld_free;
@@ -109,28 +215,381 @@  static inline bool imxdma_chan_is_doing_cyclic(struct imxdma_channel *imxdmac)
 	return false;
 }
 
-static void imxdma_irq_handler(int channel, void *data)
+/* TODO: put this inside any struct */
+static void __iomem *imx_dmav1_baseaddr;
+static struct clk *dma_clk;
+
+static void imx_dmav1_writel(unsigned val, unsigned offset)
 {
-	struct imxdma_channel *imxdmac = data;
+	__raw_writel(val, imx_dmav1_baseaddr + offset);
+}
 
-	tasklet_schedule(&imxdmac->dma_tasklet);
+static unsigned imx_dmav1_readl(unsigned offset)
+{
+	return __raw_readl(imx_dmav1_baseaddr + offset);
 }
 
-static void imxdma_err_handler(int channel, void *data, int error)
+static int imxdma_hw_chain(struct imxdma_channel_internal *imxdma)
 {
-	struct imxdma_channel *imxdmac = data;
+	if (cpu_is_mx27())
+		return imxdma->hw_chaining;
+	else
+		return 0;
+}
+
+/*
+ * imxdma_sg_next - prepare next chunk for scatter-gather DMA emulation
+ */
+static inline int imxdma_sg_next(struct imxdma_channel *imxdmac, struct scatterlist *sg)
+{
+	struct imxdma_channel_internal *imxdma = &imxdmac->internal;
+	unsigned long now;
+
+	now = min(imxdma->resbytes, sg->length);
+	if (imxdma->resbytes != IMX_DMA_LENGTH_LOOP)
+		imxdma->resbytes -= now;
+
+	if ((imxdma->dma_mode & DMA_MODE_MASK) == DMA_MODE_READ)
+		imx_dmav1_writel(sg->dma_address, DMA_DAR(imxdmac->channel));
+	else
+		imx_dmav1_writel(sg->dma_address, DMA_SAR(imxdmac->channel));
+
+	imx_dmav1_writel(now, DMA_CNTR(imxdmac->channel));
+
+	pr_debug("imxdma%d: next sg chunk dst 0x%08x, src 0x%08x, "
+		"size 0x%08x\n", imxdmac->channel,
+		 imx_dmav1_readl(DMA_DAR(imxdmac->channel)),
+		 imx_dmav1_readl(DMA_SAR(imxdmac->channel)),
+		 imx_dmav1_readl(DMA_CNTR(imxdmac->channel)));
+
+	return now;
+}
+
+static int
+imxdma_setup_single_hw(struct imxdma_channel *imxdmac, dma_addr_t dma_address,
+		     unsigned int dma_length, unsigned int dev_addr,
+		     unsigned int dmamode)
+{
+	int channel = imxdmac->channel;
+
+	imxdmac->internal.sg = NULL;
+	imxdmac->internal.dma_mode = dmamode;
+
+	if (!dma_address) {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_single null address\n",
+		       channel);
+		return -EINVAL;
+	}
+
+	if (!dma_length) {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_single zero length\n",
+		       channel);
+		return -EINVAL;
+	}
+
+	if ((dmamode & DMA_MODE_MASK) == DMA_MODE_READ) {
+		pr_debug("imxdma%d: %s dma_addressg=0x%08x dma_length=%d "
+			"dev_addr=0x%08x for read\n",
+			channel, __func__, (unsigned int)dma_address,
+			dma_length, dev_addr);
+
+		imx_dmav1_writel(dev_addr, DMA_SAR(channel));
+		imx_dmav1_writel(dma_address, DMA_DAR(channel));
+		imx_dmav1_writel(imxdmac->internal.ccr_from_device, DMA_CCR(channel));
+	} else if ((dmamode & DMA_MODE_MASK) == DMA_MODE_WRITE) {
+		pr_debug("imxdma%d: %s dma_addressg=0x%08x dma_length=%d "
+			"dev_addr=0x%08x for write\n",
+			channel, __func__, (unsigned int)dma_address,
+			dma_length, dev_addr);
+
+		imx_dmav1_writel(dma_address, DMA_SAR(channel));
+		imx_dmav1_writel(dev_addr, DMA_DAR(channel));
+		imx_dmav1_writel(imxdmac->internal.ccr_to_device,
+				DMA_CCR(channel));
+	} else {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_single bad dmamode\n",
+		       channel);
+		return -EINVAL;
+	}
 
+	imx_dmav1_writel(dma_length, DMA_CNTR(channel));
+
+	return 0;
+}
+
+static void imxdma_enable_hw(struct imxdma_channel *imxdmac)
+{
+	int channel = imxdmac->channel;
+	unsigned long flags;
+
+	pr_debug("imxdma%d: imx_dma_enable\n", channel);
+
+	if (imxdmac->internal.in_use)
+		return;
+
+	local_irq_save(flags);
+
+	imx_dmav1_writel(1 << channel, DMA_DISR);
+	imx_dmav1_writel(imx_dmav1_readl(DMA_DIMR) & ~(1 << channel), DMA_DIMR);
+	imx_dmav1_writel(imx_dmav1_readl(DMA_CCR(channel)) | CCR_CEN |
+		CCR_ACRPT, DMA_CCR(channel));
+
+	if ((cpu_is_mx21() || cpu_is_mx27()) &&
+			imxdmac->internal.sg && imxdma_hw_chain(&imxdmac->internal)) {
+		imxdmac->internal.sg = sg_next(imxdmac->internal.sg);
+		if (imxdmac->internal.sg) {
+			u32 tmp;
+			imxdma_sg_next(imxdmac, imxdmac->internal.sg);
+			tmp = imx_dmav1_readl(DMA_CCR(channel));
+			imx_dmav1_writel(tmp | CCR_RPT | CCR_ACRPT,
+				DMA_CCR(channel));
+		}
+	}
+	imxdmac->internal.in_use = 1;
+
+	local_irq_restore(flags);
+}
+
+static void imxdma_disable_hw(struct imxdma_channel *imxdmac)
+{
+	int channel = imxdmac->channel;
+	unsigned long flags;
+
+	pr_debug("imxdma%d: imx_dma_disable\n", channel);
+
+	if (imxdma_hw_chain(&imxdmac->internal))
+		del_timer(&imxdmac->internal.watchdog);
+
+	local_irq_save(flags);
+	imx_dmav1_writel(imx_dmav1_readl(DMA_DIMR) | (1 << channel), DMA_DIMR);
+	imx_dmav1_writel(imx_dmav1_readl(DMA_CCR(channel)) & ~CCR_CEN,
+			DMA_CCR(channel));
+	imx_dmav1_writel(1 << channel, DMA_DISR);
+	imxdmac->internal.in_use = 0;
+	local_irq_restore(flags);
+}
+
+static int
+imxdma_config_channel_hw(struct imxdma_channel *imxdmac, unsigned int config_port,
+	unsigned int config_mem, unsigned int dmareq, int hw_chaining)
+{
+	int channel = imxdmac->channel;
+	u32 dreq = 0;
+
+	imxdmac->internal.hw_chaining = 0;
+
+	if (hw_chaining) {
+		imxdmac->internal.hw_chaining = 1;
+		if (!imxdma_hw_chain(&imxdmac->internal))
+			return -EINVAL;
+	}
+
+	if (dmareq)
+		dreq = CCR_REN;
+
+	imxdmac->internal.ccr_from_device = config_port | (config_mem << 2) | dreq;
+	imxdmac->internal.ccr_to_device = config_mem | (config_port << 2) | dreq;
+
+	imx_dmav1_writel(dmareq, DMA_RSSR(channel));
+
+	return 0;
+}
+
+static int
+imxdma_setup_sg_hw(struct imxdma_channel *imxdmac,
+		 struct scatterlist *sg, unsigned int sgcount,
+		 unsigned int dma_length, unsigned int dev_addr,
+		 unsigned int dmamode)
+{
+	int channel = imxdmac->channel;
+
+	if (imxdmac->internal.in_use)
+		return -EBUSY;
+
+	imxdmac->internal.sg = sg;
+	imxdmac->internal.dma_mode = dmamode;
+	imxdmac->internal.resbytes = dma_length;
+
+	if (!sg || !sgcount) {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg empty sg list\n",
+		       channel);
+		return -EINVAL;
+	}
+
+	if (!sg->length) {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg zero length\n",
+		       channel);
+		return -EINVAL;
+	}
+
+	if ((dmamode & DMA_MODE_MASK) == DMA_MODE_READ) {
+		pr_debug("imxdma%d: %s sg=%p sgcount=%d total length=%d "
+			"dev_addr=0x%08x for read\n",
+			channel, __func__, sg, sgcount, dma_length, dev_addr);
+
+		imx_dmav1_writel(dev_addr, DMA_SAR(channel));
+		imx_dmav1_writel(imxdmac->internal.ccr_from_device, DMA_CCR(channel));
+	} else if ((dmamode & DMA_MODE_MASK) == DMA_MODE_WRITE) {
+		pr_debug("imxdma%d: %s sg=%p sgcount=%d total length=%d "
+			"dev_addr=0x%08x for write\n",
+			channel, __func__, sg, sgcount, dma_length, dev_addr);
+
+		imx_dmav1_writel(dev_addr, DMA_DAR(channel));
+		imx_dmav1_writel(imxdmac->internal.ccr_to_device, DMA_CCR(channel));
+	} else {
+		printk(KERN_ERR "imxdma%d: imx_dma_setup_sg bad dmamode\n",
+		       channel);
+		return -EINVAL;
+	}
+
+	imxdma_sg_next(imxdmac, sg);
+
+	return 0;
+}
+
+static void imxdma_watchdog(unsigned long data)
+{
+	struct imxdma_channel *imxdmac = (struct imxdma_channel *)data;
+	int channel = imxdmac->channel;
+
+	imx_dmav1_writel(0, DMA_CCR(channel));
+	imxdmac->internal.in_use = 0;
+	imxdmac->internal.sg = NULL;
+
+	/* Tasklet watchdog error handler */
 	tasklet_schedule(&imxdmac->dma_tasklet);
+	pr_debug("imxdma%d: watchdog timeout!\n", imxdmac->channel);
+}
+
+static irqreturn_t imxdma_err_handler(int irq, void *dev_id)
+{
+	struct imxdma_engine *imxdma = dev_id;
+	struct imxdma_channel_internal *internal;
+	unsigned int err_mask;
+	int i, disr;
+	int errcode;
+
+	disr = imx_dmav1_readl(DMA_DISR);
+
+	err_mask = imx_dmav1_readl(DMA_DBTOSR) |
+		   imx_dmav1_readl(DMA_DRTOSR) |
+		   imx_dmav1_readl(DMA_DSESR)  |
+		   imx_dmav1_readl(DMA_DBOSR);
+
+	if (!err_mask)
+		return IRQ_HANDLED;
+
+	imx_dmav1_writel(disr & err_mask, DMA_DISR);
+
+	for (i = 0; i < IMX_DMA_CHANNELS; i++) {
+		if (!(err_mask & (1 << i)))
+			continue;
+		internal = &imxdma->channel[i].internal;
+		errcode = 0;
+
+		if (imx_dmav1_readl(DMA_DBTOSR) & (1 << i)) {
+			imx_dmav1_writel(1 << i, DMA_DBTOSR);
+			errcode |= IMX_DMA_ERR_BURST;
+		}
+		if (imx_dmav1_readl(DMA_DRTOSR) & (1 << i)) {
+			imx_dmav1_writel(1 << i, DMA_DRTOSR);
+			errcode |= IMX_DMA_ERR_REQUEST;
+		}
+		if (imx_dmav1_readl(DMA_DSESR) & (1 << i)) {
+			imx_dmav1_writel(1 << i, DMA_DSESR);
+			errcode |= IMX_DMA_ERR_TRANSFER;
+		}
+		if (imx_dmav1_readl(DMA_DBOSR) & (1 << i)) {
+			imx_dmav1_writel(1 << i, DMA_DBOSR);
+			errcode |= IMX_DMA_ERR_BUFFER;
+		}
+		/* Tasklet error handler */
+		tasklet_schedule(&imxdma->channel[i].dma_tasklet);
+
+		printk(KERN_WARNING
+		       "DMA timeout on channel %d -%s%s%s%s\n", i,
+		       errcode & IMX_DMA_ERR_BURST ?    " burst" : "",
+		       errcode & IMX_DMA_ERR_REQUEST ?  " request" : "",
+		       errcode & IMX_DMA_ERR_TRANSFER ? " transfer" : "",
+		       errcode & IMX_DMA_ERR_BUFFER ?   " buffer" : "");
+	}
+	return IRQ_HANDLED;
 }
 
-static void imxdma_progression(int channel, void *data,
-		struct scatterlist *sg)
+static void dma_irq_handle_channel(struct imxdma_channel *imxdmac)
 {
-	struct imxdma_channel *imxdmac = data;
+	struct imxdma_channel_internal *imxdma = &imxdmac->internal;
+	int chno = imxdmac->channel;
+
+	if (imxdma->sg) {
+		u32 tmp;
+		imxdma->sg = sg_next(imxdma->sg);
+
+		if (imxdma->sg) {
+			imxdma_sg_next(imxdmac, imxdma->sg);
+
+			tmp = imx_dmav1_readl(DMA_CCR(chno));
+
+			if (imxdma_hw_chain(imxdma)) {
+				/* FIXME: The timeout should probably be
+				 * configurable
+				 */
+				mod_timer(&imxdma->watchdog,
+					jiffies + msecs_to_jiffies(500));
+
+				tmp |= CCR_CEN | CCR_RPT | CCR_ACRPT;
+				imx_dmav1_writel(tmp, DMA_CCR(chno));
+			} else {
+				imx_dmav1_writel(tmp & ~CCR_CEN, DMA_CCR(chno));
+				tmp |= CCR_CEN;
+			}
+
+			imx_dmav1_writel(tmp, DMA_CCR(chno));
+
+			if (imxdma_chan_is_doing_cyclic(imxdmac))
+				/* Tasklet progression */
+				tasklet_schedule(&imxdmac->dma_tasklet);
+
+			return;
+		}
 
+		if (imxdma_hw_chain(imxdma)) {
+			del_timer(&imxdma->watchdog);
+			return;
+		}
+	}
+
+	imx_dmav1_writel(0, DMA_CCR(chno));
+	imxdma->in_use = 0;
+	/* Tasklet irq */
 	tasklet_schedule(&imxdmac->dma_tasklet);
 }
 
+static irqreturn_t dma_irq_handler(int irq, void *dev_id)
+{
+	struct imxdma_engine *imxdma = dev_id;
+	struct imxdma_channel_internal *internal;
+	int i, disr;
+
+	if (cpu_is_mx21() || cpu_is_mx27())
+		imxdma_err_handler(irq, dev_id);
+
+	disr = imx_dmav1_readl(DMA_DISR);
+
+	pr_debug("imxdma: dma_irq_handler called, disr=0x%08x\n",
+		     disr);
+
+	imx_dmav1_writel(disr, DMA_DISR);
+	for (i = 0; i < IMX_DMA_CHANNELS; i++) {
+		if (disr & (1 << i)) {
+			internal = &imxdma->channel[i].internal;
+			dma_irq_handle_channel(&imxdma->channel[i]);
+		}
+	}
+
+	return IRQ_HANDLED;
+}
+
 static int imxdma_xfer_desc(struct imxdma_desc *d)
 {
 	struct imxdma_channel *imxdmac = to_imxdma_chan(d->desc.chan);
@@ -139,31 +598,24 @@  static int imxdma_xfer_desc(struct imxdma_desc *d)
 	/* Configure and enable */
 	switch (d->type) {
 	case IMXDMA_DESC_MEMCPY:
-		ret = imx_dma_config_channel(imxdmac->imxdma_channel,
+		ret = imxdma_config_channel_hw(imxdmac,
 					  d->config_port, d->config_mem, 0, 0);
 		if (ret < 0)
 			return ret;
-		ret = imx_dma_setup_single(imxdmac->imxdma_channel, d->src,
+		ret = imxdma_setup_single_hw(imxdmac, d->src,
 					   d->len, d->dest, d->dmamode);
 		if (ret < 0)
 			return ret;
 		break;
+
+	/* Cyclic transfer is the same as slave_sg with special sg configuration. */
 	case IMXDMA_DESC_CYCLIC:
-		ret = imx_dma_setup_progression_handler(imxdmac->imxdma_channel,
-							imxdma_progression);
-		if (ret < 0)
-			return ret;
-		/*
-		 * We fall through here since cyclic transfer is the same as
-		 * slave_sg adding a progression handler and a specific sg
-		 * configuration which is done in 'imxdma_prep_dma_cyclic'.
-		 */
 	case IMXDMA_DESC_SLAVE_SG:
 		if (d->dmamode == DMA_MODE_READ)
-			ret = imx_dma_setup_sg(imxdmac->imxdma_channel, d->sg,
+			ret = imxdma_setup_sg_hw(imxdmac, d->sg,
 				       d->sgcount, d->len, d->src, d->dmamode);
 		else
-			ret = imx_dma_setup_sg(imxdmac->imxdma_channel, d->sg,
+			ret = imxdma_setup_sg_hw(imxdmac, d->sg,
 				      d->sgcount, d->len, d->dest, d->dmamode);
 		if (ret < 0)
 			return ret;
@@ -171,7 +623,7 @@  static int imxdma_xfer_desc(struct imxdma_desc *d)
 	default:
 		return -EINVAL;
 	}
-	imx_dma_enable(imxdmac->imxdma_channel);
+	imxdma_enable_hw(imxdmac);
 	return 0;
 }
 
@@ -223,7 +675,7 @@  static int imxdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
 
 	switch (cmd) {
 	case DMA_TERMINATE_ALL:
-		imx_dma_disable(imxdmac->imxdma_channel);
+		imxdma_disable_hw(imxdmac);
 
 		spin_lock_irqsave(&imxdmac->lock, flags);
 		list_splice_tail_init(&imxdmac->ld_active, &imxdmac->ld_free);
@@ -253,16 +705,16 @@  static int imxdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
 			mode = IMX_DMA_MEMSIZE_32;
 			break;
 		}
-		ret = imx_dma_config_channel(imxdmac->imxdma_channel,
+		ret = imxdma_config_channel_hw(imxdmac,
 				mode | IMX_DMA_TYPE_FIFO,
 				IMX_DMA_MEMSIZE_32 | IMX_DMA_TYPE_LINEAR,
 				imxdmac->dma_request, 1);
 
 		if (ret)
 			return ret;
-
-		imx_dma_config_burstlen(imxdmac->imxdma_channel,
-				imxdmac->watermark_level * imxdmac->word_size);
+		/* Set burst length */
+		imx_dmav1_writel(imxdmac->watermark_level * imxdmac->word_size,
+				 DMA_BLR(imxdmac->channel));
 
 		return 0;
 	default:
@@ -359,7 +811,7 @@  static void imxdma_free_chan_resources(struct dma_chan *chan)
 
 	spin_lock_irqsave(&imxdmac->lock, flags);
 
-	imx_dma_disable(imxdmac->imxdma_channel);
+	imxdma_disable_hw(imxdmac);
 	list_splice_tail_init(&imxdmac->ld_active, &imxdmac->ld_free);
 	list_splice_tail_init(&imxdmac->ld_queue, &imxdmac->ld_free);
 
@@ -545,10 +997,51 @@  static void imxdma_issue_pending(struct dma_chan *chan)
 }
 
 static int __init imxdma_probe(struct platform_device *pdev)
-{
+	{
 	struct imxdma_engine *imxdma;
 	int ret, i;
 
+	if (cpu_is_mx1())
+		imx_dmav1_baseaddr = MX1_IO_ADDRESS(MX1_DMA_BASE_ADDR);
+	else if (cpu_is_mx21())
+		imx_dmav1_baseaddr = MX21_IO_ADDRESS(MX21_DMA_BASE_ADDR);
+	else if (cpu_is_mx27())
+		imx_dmav1_baseaddr = MX27_IO_ADDRESS(MX27_DMA_BASE_ADDR);
+	else
+		return 0;
+
+	dma_clk = clk_get(NULL, "dma");
+	if (IS_ERR(dma_clk))
+		return PTR_ERR(dma_clk);
+	clk_enable(dma_clk);
+
+	/* reset DMA module */
+	imx_dmav1_writel(DCR_DRST, DMA_DCR);
+
+	if (cpu_is_mx1()) {
+		ret = request_irq(MX1_DMA_INT, dma_irq_handler, 0, "DMA", imxdma);
+		if (ret) {
+			pr_crit("Can't register IRQ for DMA\n");
+			return ret;
+		}
+
+		ret = request_irq(MX1_DMA_ERR, imxdma_err_handler, 0, "DMA", imxdma);
+		if (ret) {
+			pr_crit("Can't register ERRIRQ for DMA\n");
+			free_irq(MX1_DMA_INT, NULL);
+			return ret;
+		}
+	}
+
+	/* enable DMA module */
+	imx_dmav1_writel(DCR_DEN, DMA_DCR);
+
+	/* clear all interrupts */
+	imx_dmav1_writel((1 << IMX_DMA_CHANNELS) - 1, DMA_DISR);
+
+	/* disable interrupts */
+	imx_dmav1_writel((1 << IMX_DMA_CHANNELS) - 1, DMA_DIMR);
+
 	imxdma = kzalloc(sizeof(*imxdma), GFP_KERNEL);
 	if (!imxdma)
 		return -ENOMEM;
@@ -562,17 +1055,20 @@  static int __init imxdma_probe(struct platform_device *pdev)
 	/* Initialize channel parameters */
 	for (i = 0; i < MAX_DMA_CHANNELS; i++) {
 		struct imxdma_channel *imxdmac = &imxdma->channel[i];
-
-		imxdmac->imxdma_channel = imx_dma_request_by_prio("dmaengine",
-				DMA_PRIO_MEDIUM);
-		if ((int)imxdmac->channel < 0) {
-			ret = -ENODEV;
-			goto err_init;
+		memset(&imxdmac->internal, 0, sizeof(imxdmac->internal));
+		if (cpu_is_mx21() || cpu_is_mx27()) {
+			ret = request_irq(MX2x_INT_DMACH0 + i,
+					dma_irq_handler, 0, "DMA", imxdma);
+			if (ret) {
+				pr_crit("Can't register IRQ %d for DMA channel %d\n",
+						MX2x_INT_DMACH0 + i, i);
+				goto err_init;
+			}
+			init_timer(&imxdmac->internal.watchdog);
+			imxdmac->internal.watchdog.function = &imxdma_watchdog;
+			imxdmac->internal.watchdog.data = (unsigned long)imxdmac;
 		}
 
-		imx_dma_setup_handlers(imxdmac->imxdma_channel,
-		       imxdma_irq_handler, imxdma_err_handler, imxdmac);
-
 		imxdmac->imxdma = imxdma;
 		spin_lock_init(&imxdmac->lock);
 
@@ -617,10 +1113,8 @@  static int __init imxdma_probe(struct platform_device *pdev)
 	return 0;
 
 err_init:
-	while (--i >= 0) {
-		struct imxdma_channel *imxdmac = &imxdma->channel[i];
-		imx_dma_free(imxdmac->imxdma_channel);
-	}
+	while (--i >= 0)
+		free_irq(MX2x_INT_DMACH0 + i, NULL);
 
 	kfree(imxdma);
 	return ret;
@@ -633,11 +1127,8 @@  static int __exit imxdma_remove(struct platform_device *pdev)
 
         dma_async_device_unregister(&imxdma->dma_device);
 
-	for (i = 0; i < MAX_DMA_CHANNELS; i++) {
-		struct imxdma_channel *imxdmac = &imxdma->channel[i];
-
-		 imx_dma_free(imxdmac->imxdma_channel);
-	}
+	for (i = 0; i < MAX_DMA_CHANNELS; i++)
+		free_irq(MX2x_INT_DMACH0 + i, NULL);
 
         kfree(imxdma);