Merge tag 'dmaengine-4.2-rc1' of git://git.infradead.org/users/vkoul/slave-dma

Pull dmaengine updates from Vinod Koul:
 "This time we have support for few new devices, few new features and
  odd fixes spread thru the subsystem.

  New devices added:
   - support for CSRatlas7 dma controller
   - Allwinner H3(sun8i) controller
   - TI DMA crossbar driver on DRA7x
   - new pxa driver

  New features added:
   - memset support is bought back now that we have a user in xdmac controller
   - interleaved transfers support different source and destination strides
   - supporting DMA routers and configuration thru DT
   - support for reusing descriptors
   - xdmac memset and interleaved transfer support
   - hdmac support for interleaved transfers
   - omap-dma support for memcpy

  Others:
   - Constify platform_device_id
   - mv_xor fixes and improvements"

* tag 'dmaengine-4.2-rc1' of git://git.infradead.org/users/vkoul/slave-dma: (46 commits)
  dmaengine: xgene: fix file permission
  dmaengine: fsl-edma: clear pending interrupts on initialization
  dmaengine: xdmac: Add memset support
  Documentation: dmaengine: document DMA_CTRL_ACK
  dmaengine: virt-dma: don't always free descriptor upon completion
  dmaengine: Revert "drivers/dma: remove unused support for MEMSET operations"
  dmaengine: hdmac: Implement interleaved transfers
  dmaengine: Move icg helpers to global header
  dmaengine: mv_xor: improve descriptors list handling and reduce locking
  dmaengine: mv_xor: Enlarge descriptor pool size
  dmaengine: mv_xor: add support for a38x command in descriptor mode
  dmaengine: mv_xor: Rename function for consistent naming
  dmaengine: mv_xor: bug fix for racing condition in descriptors cleanup
  dmaengine: pl330: fix wording in mcbufsz message
  dmaengine: sirf: add CSRatlas7 SoC support
  dmaengine: xgene-dma: Fix "incorrect type in assignement" warnings
  dmaengine: fix kernel-doc documentation
  dmaengine: pxa_dma: add support for legacy transition
  dmaengine: pxa_dma: add debug information
  dmaengine: pxa: add pxa dmaengine driver
  ...

1 files changed, 334 insertions, 89 deletions

diff --git a/drivers/dma/sirf-dma.c b/drivers/dma/sirf-dma.c
index a1afda43b8ef..8c5186cc9f63 100644
--- a/drivers/dma/sirf-dma.c
+++ b/drivers/dma/sirf-dma.c
@@ -23,8 +23,13 @@
 
 #include "dmaengine.h"
 
+#define SIRFSOC_DMA_VER_A7V1                    1
+#define SIRFSOC_DMA_VER_A7V2                    2
+#define SIRFSOC_DMA_VER_A6                      4
+
 #define SIRFSOC_DMA_DESCRIPTORS                 16
 #define SIRFSOC_DMA_CHANNELS                    16
+#define SIRFSOC_DMA_TABLE_NUM                   256
 
 #define SIRFSOC_DMA_CH_ADDR                     0x00
 #define SIRFSOC_DMA_CH_XLEN                     0x04
@@ -35,15 +40,44 @@
 #define SIRFSOC_DMA_CH_VALID                    0x140
 #define SIRFSOC_DMA_CH_INT                      0x144
 #define SIRFSOC_DMA_INT_EN                      0x148
-#define SIRFSOC_DMA_INT_EN_CLR			0x14C
+#define SIRFSOC_DMA_INT_EN_CLR                  0x14C
 #define SIRFSOC_DMA_CH_LOOP_CTRL                0x150
-#define SIRFSOC_DMA_CH_LOOP_CTRL_CLR            0x15C
+#define SIRFSOC_DMA_CH_LOOP_CTRL_CLR            0x154
+#define SIRFSOC_DMA_WIDTH_ATLAS7                0x10
+#define SIRFSOC_DMA_VALID_ATLAS7                0x14
+#define SIRFSOC_DMA_INT_ATLAS7                  0x18
+#define SIRFSOC_DMA_INT_EN_ATLAS7               0x1c
+#define SIRFSOC_DMA_LOOP_CTRL_ATLAS7            0x20
+#define SIRFSOC_DMA_CUR_DATA_ADDR               0x34
+#define SIRFSOC_DMA_MUL_ATLAS7                  0x38
+#define SIRFSOC_DMA_CH_LOOP_CTRL_ATLAS7         0x158
+#define SIRFSOC_DMA_CH_LOOP_CTRL_CLR_ATLAS7     0x15C
+#define SIRFSOC_DMA_IOBG_SCMD_EN		0x800
+#define SIRFSOC_DMA_EARLY_RESP_SET		0x818
+#define SIRFSOC_DMA_EARLY_RESP_CLR		0x81C
 
 #define SIRFSOC_DMA_MODE_CTRL_BIT               4
 #define SIRFSOC_DMA_DIR_CTRL_BIT                5
+#define SIRFSOC_DMA_MODE_CTRL_BIT_ATLAS7        2
+#define SIRFSOC_DMA_CHAIN_CTRL_BIT_ATLAS7       3
+#define SIRFSOC_DMA_DIR_CTRL_BIT_ATLAS7         4
+#define SIRFSOC_DMA_TAB_NUM_ATLAS7              7
+#define SIRFSOC_DMA_CHAIN_INT_BIT_ATLAS7        5
+#define SIRFSOC_DMA_CHAIN_FLAG_SHIFT_ATLAS7     25
+#define SIRFSOC_DMA_CHAIN_ADDR_SHIFT            32
+
+#define SIRFSOC_DMA_INT_FINI_INT_ATLAS7         BIT(0)
+#define SIRFSOC_DMA_INT_CNT_INT_ATLAS7          BIT(1)
+#define SIRFSOC_DMA_INT_PAU_INT_ATLAS7          BIT(2)
+#define SIRFSOC_DMA_INT_LOOP_INT_ATLAS7         BIT(3)
+#define SIRFSOC_DMA_INT_INV_INT_ATLAS7          BIT(4)
+#define SIRFSOC_DMA_INT_END_INT_ATLAS7          BIT(5)
+#define SIRFSOC_DMA_INT_ALL_ATLAS7              0x3F
 
 /* xlen and dma_width register is in 4 bytes boundary */
 #define SIRFSOC_DMA_WORD_LEN			4
+#define SIRFSOC_DMA_XLEN_MAX_V1         0x800
+#define SIRFSOC_DMA_XLEN_MAX_V2         0x1000
 
 struct sirfsoc_dma_desc {
 	struct dma_async_tx_descriptor	desc;
@@ -56,7 +90,9 @@ struct sirfsoc_dma_desc {
 	int             width;          /* DMA width */
 	int             dir;
 	bool            cyclic;         /* is loop DMA? */
+	bool            chain;          /* is chain DMA? */
 	u32             addr;		/* DMA buffer address */
+	u64 chain_table[SIRFSOC_DMA_TABLE_NUM]; /* chain tbl */
 };
 
 struct sirfsoc_dma_chan {
@@ -87,10 +123,25 @@ struct sirfsoc_dma {
 	void __iomem			*base;
 	int				irq;
 	struct clk			*clk;
-	bool				is_marco;
+	int				type;
+	void (*exec_desc)(struct sirfsoc_dma_desc *sdesc,
+		int cid, int burst_mode, void __iomem *base);
 	struct sirfsoc_dma_regs		regs_save;
 };
 
+struct sirfsoc_dmadata {
+	void (*exec)(struct sirfsoc_dma_desc *sdesc,
+		int cid, int burst_mode, void __iomem *base);
+	int type;
+};
+
+enum sirfsoc_dma_chain_flag {
+	SIRFSOC_DMA_CHAIN_NORMAL = 0x01,
+	SIRFSOC_DMA_CHAIN_PAUSE = 0x02,
+	SIRFSOC_DMA_CHAIN_LOOP = 0x03,
+	SIRFSOC_DMA_CHAIN_END = 0x04
+};
+
 #define DRV_NAME	"sirfsoc_dma"
 
 static int sirfsoc_dma_runtime_suspend(struct device *dev);
@@ -109,48 +160,105 @@ static inline struct sirfsoc_dma *dma_chan_to_sirfsoc_dma(struct dma_chan *c)
 	return container_of(schan, struct sirfsoc_dma, channels[c->chan_id]);
 }
 
+static void sirfsoc_dma_execute_hw_a7v2(struct sirfsoc_dma_desc *sdesc,
+		int cid, int burst_mode, void __iomem *base)
+{
+	if (sdesc->chain) {
+		/* DMA v2 HW chain mode */
+		writel_relaxed((sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT_ATLAS7) |
+			       (sdesc->chain <<
+				SIRFSOC_DMA_CHAIN_CTRL_BIT_ATLAS7) |
+			       (0x8 << SIRFSOC_DMA_TAB_NUM_ATLAS7) | 0x3,
+			       base + SIRFSOC_DMA_CH_CTRL);
+	} else {
+		/* DMA v2 legacy mode */
+		writel_relaxed(sdesc->xlen, base + SIRFSOC_DMA_CH_XLEN);
+		writel_relaxed(sdesc->ylen, base + SIRFSOC_DMA_CH_YLEN);
+		writel_relaxed(sdesc->width, base + SIRFSOC_DMA_WIDTH_ATLAS7);
+		writel_relaxed((sdesc->width*((sdesc->ylen+1)>>1)),
+				base + SIRFSOC_DMA_MUL_ATLAS7);
+		writel_relaxed((sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT_ATLAS7) |
+			       (sdesc->chain <<
+				SIRFSOC_DMA_CHAIN_CTRL_BIT_ATLAS7) |
+			       0x3, base + SIRFSOC_DMA_CH_CTRL);
+	}
+	writel_relaxed(sdesc->chain ? SIRFSOC_DMA_INT_END_INT_ATLAS7 :
+		       (SIRFSOC_DMA_INT_FINI_INT_ATLAS7 |
+			SIRFSOC_DMA_INT_LOOP_INT_ATLAS7),
+		       base + SIRFSOC_DMA_INT_EN_ATLAS7);
+	writel(sdesc->addr, base + SIRFSOC_DMA_CH_ADDR);
+	if (sdesc->cyclic)
+		writel(0x10001, base + SIRFSOC_DMA_LOOP_CTRL_ATLAS7);
+}
+
+static void sirfsoc_dma_execute_hw_a7v1(struct sirfsoc_dma_desc *sdesc,
+		int cid, int burst_mode, void __iomem *base)
+{
+	writel_relaxed(1, base + SIRFSOC_DMA_IOBG_SCMD_EN);
+	writel_relaxed((1 << cid), base + SIRFSOC_DMA_EARLY_RESP_SET);
+	writel_relaxed(sdesc->width, base + SIRFSOC_DMA_WIDTH_0 + cid * 4);
+	writel_relaxed(cid | (burst_mode << SIRFSOC_DMA_MODE_CTRL_BIT) |
+		       (sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT),
+		       base + cid * 0x10 + SIRFSOC_DMA_CH_CTRL);
+	writel_relaxed(sdesc->xlen, base + cid * 0x10 + SIRFSOC_DMA_CH_XLEN);
+	writel_relaxed(sdesc->ylen, base + cid * 0x10 + SIRFSOC_DMA_CH_YLEN);
+	writel_relaxed(readl_relaxed(base + SIRFSOC_DMA_INT_EN) |
+		       (1 << cid), base + SIRFSOC_DMA_INT_EN);
+	writel(sdesc->addr >> 2, base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR);
+	if (sdesc->cyclic) {
+		writel((1 << cid) | 1 << (cid + 16) |
+		       readl_relaxed(base + SIRFSOC_DMA_CH_LOOP_CTRL_ATLAS7),
+		       base + SIRFSOC_DMA_CH_LOOP_CTRL_ATLAS7);
+	}
+
+}
+
+static void sirfsoc_dma_execute_hw_a6(struct sirfsoc_dma_desc *sdesc,
+		int cid, int burst_mode, void __iomem *base)
+{
+	writel_relaxed(sdesc->width, base + SIRFSOC_DMA_WIDTH_0 + cid * 4);
+	writel_relaxed(cid | (burst_mode << SIRFSOC_DMA_MODE_CTRL_BIT) |
+		       (sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT),
+		       base + cid * 0x10 + SIRFSOC_DMA_CH_CTRL);
+	writel_relaxed(sdesc->xlen, base + cid * 0x10 + SIRFSOC_DMA_CH_XLEN);
+	writel_relaxed(sdesc->ylen, base + cid * 0x10 + SIRFSOC_DMA_CH_YLEN);
+	writel_relaxed(readl_relaxed(base + SIRFSOC_DMA_INT_EN) |
+		       (1 << cid), base + SIRFSOC_DMA_INT_EN);
+	writel(sdesc->addr >> 2, base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR);
+	if (sdesc->cyclic) {
+		writel((1 << cid) | 1 << (cid + 16) |
+		       readl_relaxed(base + SIRFSOC_DMA_CH_LOOP_CTRL),
+		       base + SIRFSOC_DMA_CH_LOOP_CTRL);
+	}
+
+}
+
 /* Execute all queued DMA descriptors */
 static void sirfsoc_dma_execute(struct sirfsoc_dma_chan *schan)
 {
 	struct sirfsoc_dma *sdma = dma_chan_to_sirfsoc_dma(&schan->chan);
 	int cid = schan->chan.chan_id;
 	struct sirfsoc_dma_desc *sdesc = NULL;
+	void __iomem *base;
 
 	/*
 	 * lock has been held by functions calling this, so we don't hold
 	 * lock again
 	 */
-
+	base = sdma->base;
 	sdesc = list_first_entry(&schan->queued, struct sirfsoc_dma_desc,
-		node);
+				 node);
 	/* Move the first queued descriptor to active list */
 	list_move_tail(&sdesc->node, &schan->active);
 
-	/* Start the DMA transfer */
-	writel_relaxed(sdesc->width, sdma->base + SIRFSOC_DMA_WIDTH_0 +
-		cid * 4);
-	writel_relaxed(cid | (schan->mode << SIRFSOC_DMA_MODE_CTRL_BIT) |
-		(sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT),
-		sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_CTRL);
-	writel_relaxed(sdesc->xlen, sdma->base + cid * 0x10 +
-		SIRFSOC_DMA_CH_XLEN);
-	writel_relaxed(sdesc->ylen, sdma->base + cid * 0x10 +
-		SIRFSOC_DMA_CH_YLEN);
-	writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN) |
-		(1 << cid), sdma->base + SIRFSOC_DMA_INT_EN);
+	if (sdma->type == SIRFSOC_DMA_VER_A7V2)
+		cid = 0;
 
-	/*
-	 * writel has an implict memory write barrier to make sure data is
-	 * flushed into memory before starting DMA
-	 */
-	writel(sdesc->addr >> 2, sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR);
+	/* Start the DMA transfer */
+	sdma->exec_desc(sdesc, cid, schan->mode, base);
 
-	if (sdesc->cyclic) {
-		writel((1 << cid) | 1 << (cid + 16) |
-			readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+	if (sdesc->cyclic)
 		schan->happened_cyclic = schan->completed_cyclic = 0;
-	}
 }
 
 /* Interrupt handler */
@@ -160,27 +268,65 @@ static irqreturn_t sirfsoc_dma_irq(int irq, void *data)
 	struct sirfsoc_dma_chan *schan;
 	struct sirfsoc_dma_desc *sdesc = NULL;
 	u32 is;
+	bool chain;
 	int ch;
+	void __iomem *reg;
+
+	switch (sdma->type) {
+	case SIRFSOC_DMA_VER_A6:
+	case SIRFSOC_DMA_VER_A7V1:
+		is = readl(sdma->base + SIRFSOC_DMA_CH_INT);
+		reg = sdma->base + SIRFSOC_DMA_CH_INT;
+		while ((ch = fls(is) - 1) >= 0) {
+			is &= ~(1 << ch);
+			writel_relaxed(1 << ch, reg);
+			schan = &sdma->channels[ch];
+			spin_lock(&schan->lock);
+			sdesc = list_first_entry(&schan->active,
+						 struct sirfsoc_dma_desc, node);
+			if (!sdesc->cyclic) {
+				/* Execute queued descriptors */
+				list_splice_tail_init(&schan->active,
+						      &schan->completed);
+				dma_cookie_complete(&sdesc->desc);
+				if (!list_empty(&schan->queued))
+					sirfsoc_dma_execute(schan);
+			} else
+				schan->happened_cyclic++;
+			spin_unlock(&schan->lock);
+		}
+		break;
 
-	is = readl(sdma->base + SIRFSOC_DMA_CH_INT);
-	while ((ch = fls(is) - 1) >= 0) {
-		is &= ~(1 << ch);
-		writel_relaxed(1 << ch, sdma->base + SIRFSOC_DMA_CH_INT);
-		schan = &sdma->channels[ch];
+	case SIRFSOC_DMA_VER_A7V2:
+		is = readl(sdma->base + SIRFSOC_DMA_INT_ATLAS7);
 
+		reg = sdma->base + SIRFSOC_DMA_INT_ATLAS7;
+		writel_relaxed(SIRFSOC_DMA_INT_ALL_ATLAS7, reg);
+		schan = &sdma->channels[0];
 		spin_lock(&schan->lock);
-
-		sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc,
-			node);
+		sdesc = list_first_entry(&schan->active,
+					 struct sirfsoc_dma_desc, node);
 		if (!sdesc->cyclic) {
-			/* Execute queued descriptors */
-			list_splice_tail_init(&schan->active, &schan->completed);
-			if (!list_empty(&schan->queued))
-				sirfsoc_dma_execute(schan);
-		} else
+			chain = sdesc->chain;
+			if ((chain && (is & SIRFSOC_DMA_INT_END_INT_ATLAS7)) ||
+				(!chain &&
+				(is & SIRFSOC_DMA_INT_FINI_INT_ATLAS7))) {
+				/* Execute queued descriptors */
+				list_splice_tail_init(&schan->active,
+						      &schan->completed);
+				dma_cookie_complete(&sdesc->desc);
+				if (!list_empty(&schan->queued))
+					sirfsoc_dma_execute(schan);
+			}
+		} else if (sdesc->cyclic && (is &
+					SIRFSOC_DMA_INT_LOOP_INT_ATLAS7))
 			schan->happened_cyclic++;
 
 		spin_unlock(&schan->lock);
+		break;
+
+	default:
+		break;
 	}
 
 	/* Schedule tasklet */
@@ -227,16 +373,15 @@ static void sirfsoc_dma_process_completed(struct sirfsoc_dma *sdma)
 			schan->chan.completed_cookie = last_cookie;
 			spin_unlock_irqrestore(&schan->lock, flags);
 		} else {
-			/* for cyclic channel, desc is always in active list */
-			sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc,
-				node);
-
-			if (!sdesc || (sdesc && !sdesc->cyclic)) {
-				/* without active cyclic DMA */
+			if (list_empty(&schan->active)) {
 				spin_unlock_irqrestore(&schan->lock, flags);
 				continue;
 			}
 
+			/* for cyclic channel, desc is always in active list */
+			sdesc = list_first_entry(&schan->active,
+				struct sirfsoc_dma_desc, node);
+
 			/* cyclic DMA */
 			happened_cyclic = schan->happened_cyclic;
 			spin_unlock_irqrestore(&schan->lock, flags);
@@ -307,20 +452,32 @@ static int sirfsoc_dma_terminate_all(struct dma_chan *chan)
 
 	spin_lock_irqsave(&schan->lock, flags);
 
-	if (!sdma->is_marco) {
-		writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN) &
-			~(1 << cid), sdma->base + SIRFSOC_DMA_INT_EN);
-		writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL)
-			& ~((1 << cid) | 1 << (cid + 16)),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
-	} else {
+	switch (sdma->type) {
+	case SIRFSOC_DMA_VER_A7V1:
 		writel_relaxed(1 << cid, sdma->base + SIRFSOC_DMA_INT_EN_CLR);
 		writel_relaxed((1 << cid) | 1 << (cid + 16),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL_CLR);
+			       sdma->base +
+			       SIRFSOC_DMA_CH_LOOP_CTRL_CLR_ATLAS7);
+		writel_relaxed(1 << cid, sdma->base + SIRFSOC_DMA_CH_VALID);
+		break;
+	case SIRFSOC_DMA_VER_A7V2:
+		writel_relaxed(0, sdma->base + SIRFSOC_DMA_INT_EN_ATLAS7);
+		writel_relaxed(0, sdma->base + SIRFSOC_DMA_LOOP_CTRL_ATLAS7);
+		writel_relaxed(0, sdma->base + SIRFSOC_DMA_VALID_ATLAS7);
+		break;
+	case SIRFSOC_DMA_VER_A6:
+		writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN) &
+			       ~(1 << cid), sdma->base + SIRFSOC_DMA_INT_EN);
+		writel_relaxed(readl_relaxed(sdma->base +
+					     SIRFSOC_DMA_CH_LOOP_CTRL) &
+			       ~((1 << cid) | 1 << (cid + 16)),
+			       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+		writel_relaxed(1 << cid, sdma->base + SIRFSOC_DMA_CH_VALID);
+		break;
+	default:
+		break;
 	}
 
-	writel_relaxed(1 << cid, sdma->base + SIRFSOC_DMA_CH_VALID);
-
 	list_splice_tail_init(&schan->active, &schan->free);
 	list_splice_tail_init(&schan->queued, &schan->free);
 
@@ -338,13 +495,25 @@ static int sirfsoc_dma_pause_chan(struct dma_chan *chan)
 
 	spin_lock_irqsave(&schan->lock, flags);
 
-	if (!sdma->is_marco)
-		writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL)
-			& ~((1 << cid) | 1 << (cid + 16)),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
-	else
+	switch (sdma->type) {
+	case SIRFSOC_DMA_VER_A7V1:
 		writel_relaxed((1 << cid) | 1 << (cid + 16),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL_CLR);
+			       sdma->base +
+			       SIRFSOC_DMA_CH_LOOP_CTRL_CLR_ATLAS7);
+		break;
+	case SIRFSOC_DMA_VER_A7V2:
+		writel_relaxed(0, sdma->base + SIRFSOC_DMA_LOOP_CTRL_ATLAS7);
+		break;
+	case SIRFSOC_DMA_VER_A6:
+		writel_relaxed(readl_relaxed(sdma->base +
+					     SIRFSOC_DMA_CH_LOOP_CTRL) &
+			       ~((1 << cid) | 1 << (cid + 16)),
+			       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+		break;
+
+	default:
+		break;
+	}
 
 	spin_unlock_irqrestore(&schan->lock, flags);
 
@@ -359,14 +528,25 @@ static int sirfsoc_dma_resume_chan(struct dma_chan *chan)
 	unsigned long flags;
 
 	spin_lock_irqsave(&schan->lock, flags);
-
-	if (!sdma->is_marco)
-		writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL)
-			| ((1 << cid) | 1 << (cid + 16)),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
-	else
+	switch (sdma->type) {
+	case SIRFSOC_DMA_VER_A7V1:
 		writel_relaxed((1 << cid) | 1 << (cid + 16),
-			sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+			       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL_ATLAS7);
+		break;
+	case SIRFSOC_DMA_VER_A7V2:
+		writel_relaxed(0x10001,
+			       sdma->base + SIRFSOC_DMA_LOOP_CTRL_ATLAS7);
+		break;
+	case SIRFSOC_DMA_VER_A6:
+		writel_relaxed(readl_relaxed(sdma->base +
+					     SIRFSOC_DMA_CH_LOOP_CTRL) |
+			       ((1 << cid) | 1 << (cid + 16)),
+			       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+		break;
+
+	default:
+		break;
+	}
 
 	spin_unlock_irqrestore(&schan->lock, flags);
 
@@ -473,14 +653,31 @@ sirfsoc_dma_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 
 	spin_lock_irqsave(&schan->lock, flags);
 
-	sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc,
-			node);
-	dma_request_bytes = (sdesc->xlen + 1) * (sdesc->ylen + 1) *
-		(sdesc->width * SIRFSOC_DMA_WORD_LEN);
+	if (list_empty(&schan->active)) {
+		ret = dma_cookie_status(chan, cookie, txstate);
+		dma_set_residue(txstate, 0);
+		spin_unlock_irqrestore(&schan->lock, flags);
+		return ret;
+	}
+	sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc, node);
+	if (sdesc->cyclic)
+		dma_request_bytes = (sdesc->xlen + 1) * (sdesc->ylen + 1) *
+			(sdesc->width * SIRFSOC_DMA_WORD_LEN);
+	else
+		dma_request_bytes = sdesc->xlen * SIRFSOC_DMA_WORD_LEN;
 
 	ret = dma_cookie_status(chan, cookie, txstate);
-	dma_pos = readl_relaxed(sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR)
-		<< 2;
+
+	if (sdma->type == SIRFSOC_DMA_VER_A7V2)
+		cid = 0;
+
+	if (sdma->type == SIRFSOC_DMA_VER_A7V2) {
+		dma_pos = readl_relaxed(sdma->base + SIRFSOC_DMA_CUR_DATA_ADDR);
+	} else {
+		dma_pos = readl_relaxed(
+			sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR) << 2;
+	}
+
 	residue = dma_request_bytes - (dma_pos - sdesc->addr);
 	dma_set_residue(txstate, residue);
 
@@ -647,6 +844,7 @@ static int sirfsoc_dma_probe(struct platform_device *op)
 	struct dma_device *dma;
 	struct sirfsoc_dma *sdma;
 	struct sirfsoc_dma_chan *schan;
+	struct sirfsoc_dmadata *data;
 	struct resource res;
 	ulong regs_start, regs_size;
 	u32 id;
@@ -657,9 +855,11 @@ static int sirfsoc_dma_probe(struct platform_device *op)
 		dev_err(dev, "Memory exhausted!\n");
 		return -ENOMEM;
 	}
-
-	if (of_device_is_compatible(dn, "sirf,marco-dmac"))
-		sdma->is_marco = true;
+	data = (struct sirfsoc_dmadata *)
+		(of_match_device(op->dev.driver->of_match_table,
+				 &op->dev)->data);
+	sdma->exec_desc = data->exec;
+	sdma->type = data->type;
 
 	if (of_property_read_u32(dn, "cell-index", &id)) {
 		dev_err(dev, "Fail to get DMAC index\n");
@@ -816,6 +1016,8 @@ static int sirfsoc_dma_pm_suspend(struct device *dev)
 	struct sirfsoc_dma_chan *schan;
 	int ch;
 	int ret;
+	int count;
+	u32 int_offset;
 
 	/*
 	 * if we were runtime-suspended before, resume to enable clock
@@ -827,11 +1029,19 @@ static int sirfsoc_dma_pm_suspend(struct device *dev)
 			return ret;
 	}
 
+	if (sdma->type == SIRFSOC_DMA_VER_A7V2) {
+		count = 1;
+		int_offset = SIRFSOC_DMA_INT_EN_ATLAS7;
+	} else {
+		count = SIRFSOC_DMA_CHANNELS;
+		int_offset = SIRFSOC_DMA_INT_EN;
+	}
+
 	/*
 	 * DMA controller will lose all registers while suspending
 	 * so we need to save registers for active channels
 	 */
-	for (ch = 0; ch < SIRFSOC_DMA_CHANNELS; ch++) {
+	for (ch = 0; ch < count; ch++) {
 		schan = &sdma->channels[ch];
 		if (list_empty(&schan->active))
 			continue;
@@ -841,7 +1051,7 @@ static int sirfsoc_dma_pm_suspend(struct device *dev)
 		save->ctrl[ch] = readl_relaxed(sdma->base +
 			ch * 0x10 + SIRFSOC_DMA_CH_CTRL);
 	}
-	save->interrupt_en = readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN);
+	save->interrupt_en = readl_relaxed(sdma->base + int_offset);
 
 	/* Disable clock */
 	sirfsoc_dma_runtime_suspend(dev);
@@ -857,14 +1067,27 @@ static int sirfsoc_dma_pm_resume(struct device *dev)
 	struct sirfsoc_dma_chan *schan;
 	int ch;
 	int ret;
+	int count;
+	u32 int_offset;
+	u32 width_offset;
 
 	/* Enable clock before accessing register */
 	ret = sirfsoc_dma_runtime_resume(dev);
 	if (ret < 0)
 		return ret;
 
-	writel_relaxed(save->interrupt_en, sdma->base + SIRFSOC_DMA_INT_EN);
-	for (ch = 0; ch < SIRFSOC_DMA_CHANNELS; ch++) {
+	if (sdma->type == SIRFSOC_DMA_VER_A7V2) {
+		count = 1;
+		int_offset = SIRFSOC_DMA_INT_EN_ATLAS7;
+		width_offset = SIRFSOC_DMA_WIDTH_ATLAS7;
+	} else {
+		count = SIRFSOC_DMA_CHANNELS;
+		int_offset = SIRFSOC_DMA_INT_EN;
+		width_offset = SIRFSOC_DMA_WIDTH_0;
+	}
+
+	writel_relaxed(save->interrupt_en, sdma->base + int_offset);
+	for (ch = 0; ch < count; ch++) {
 		schan = &sdma->channels[ch];
 		if (list_empty(&schan->active))
 			continue;
@@ -872,15 +1095,21 @@ static int sirfsoc_dma_pm_resume(struct device *dev)
 			struct sirfsoc_dma_desc,
 			node);
 		writel_relaxed(sdesc->width,
-			sdma->base + SIRFSOC_DMA_WIDTH_0 + ch * 4);
+			sdma->base + width_offset + ch * 4);
 		writel_relaxed(sdesc->xlen,
 			sdma->base + ch * 0x10 + SIRFSOC_DMA_CH_XLEN);
 		writel_relaxed(sdesc->ylen,
 			sdma->base + ch * 0x10 + SIRFSOC_DMA_CH_YLEN);
 		writel_relaxed(save->ctrl[ch],
 			sdma->base + ch * 0x10 + SIRFSOC_DMA_CH_CTRL);
-		writel_relaxed(sdesc->addr >> 2,
-			sdma->base + ch * 0x10 + SIRFSOC_DMA_CH_ADDR);
+		if (sdma->type == SIRFSOC_DMA_VER_A7V2) {
+			writel_relaxed(sdesc->addr,
+				sdma->base + SIRFSOC_DMA_CH_ADDR);
+		} else {
+			writel_relaxed(sdesc->addr >> 2,
+				sdma->base + ch * 0x10 + SIRFSOC_DMA_CH_ADDR);
+
+		}
 	}
 
 	/* if we were runtime-suspended before, suspend again */
@@ -896,9 +1125,25 @@ static const struct dev_pm_ops sirfsoc_dma_pm_ops = {
 	SET_SYSTEM_SLEEP_PM_OPS(sirfsoc_dma_pm_suspend, sirfsoc_dma_pm_resume)
 };
 
+struct sirfsoc_dmadata sirfsoc_dmadata_a6 = {
+	.exec = sirfsoc_dma_execute_hw_a6,
+	.type = SIRFSOC_DMA_VER_A6,
+};
+
+struct sirfsoc_dmadata sirfsoc_dmadata_a7v1 = {
+	.exec = sirfsoc_dma_execute_hw_a7v1,
+	.type = SIRFSOC_DMA_VER_A7V1,
+};
+
+struct sirfsoc_dmadata sirfsoc_dmadata_a7v2 = {
+	.exec = sirfsoc_dma_execute_hw_a7v2,
+	.type = SIRFSOC_DMA_VER_A7V2,
+};
+
 static const struct of_device_id sirfsoc_dma_match[] = {
-	{ .compatible = "sirf,prima2-dmac", },
-	{ .compatible = "sirf,marco-dmac", },
+	{ .compatible = "sirf,prima2-dmac", .data = &sirfsoc_dmadata_a6,},
+	{ .compatible = "sirf,atlas7-dmac", .data = &sirfsoc_dmadata_a7v1,},
+	{ .compatible = "sirf,atlas7-dmac-v2", .data = &sirfsoc_dmadata_a7v2,},
 	{},
 };
 
@@ -925,7 +1170,7 @@ static void __exit sirfsoc_dma_exit(void)
 subsys_initcall(sirfsoc_dma_init);
 module_exit(sirfsoc_dma_exit);
 
-MODULE_AUTHOR("Rongjun Ying <rongjun.ying@csr.com>, "
-	"Barry Song <baohua.song@csr.com>");
+MODULE_AUTHOR("Rongjun Ying <rongjun.ying@csr.com>");
+MODULE_AUTHOR("Barry Song <baohua.song@csr.com>");
 MODULE_DESCRIPTION("SIRFSOC DMA control driver");
 MODULE_LICENSE("GPL v2");