[PATCH 0/8 v4] bcm2835 DMA slave support

[PATCH 0/8 v4] bcm2835 DMA slave support

[Eric Anholt]

Here's the series for DMA slave and memcpy support for 2835, with the
DT changes to enable the remaining channels dropped out while that
goes through review.  I had to do some minor conflict resolution, but
it was pretty mechanical, and I tested again with dmatest on the last
patch.

Martin Sperl (8):
  dmaengine: bcm2835: set residue_granularity field
  dmaengine: bcm2835: remove unnecessary masking of dma channels
  dmaengine: bcm2835: add additional defines for DMA-registers
  dmaengine: bcm2835: move cyclic member from bcm2835_chan into
    bcm2835_desc
  dmaengine: bcm2835: move controlblock chain generation into separate
    method
  dmaengine: bcm2835: limit max length based on channel type
  dmaengine: bcm2835: add slave_sg support to bcm2835-dma
  dmaengine: bcm2835: add dma_memcopy support to bcm2835-dma

 drivers/dma/bcm2835-dma.c | 529 ++++++++++++++++++++++++++++++++++++----------
 1 file changed, 412 insertions(+), 117 deletions(-)

-- 
2.7.0

[PATCH 1/8] dmaengine: bcm2835: set residue_granularity field

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

bcm2835-dma supports residue reporting at burst level but didn't report
this via the residue_granularity field.

See also:
https://github.com/raspberrypi/linux/commit/b015555327afa402f70ddc86e3632f59df1cd9d7
for the downstream patch.

Signed-off-by: Matthias Reichl <hias@horus.com>
Signed-off-by: Noralf Tr?nnes <noralf@tronnes.org>
Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index 996c4b0..2d72fe8 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -625,6 +625,7 @@ static int bcm2835_dma_probe(struct platform_device *pdev)
 	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
 	od->ddev.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
 	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
+	od->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
 	od->ddev.dev = &pdev->dev;
 	INIT_LIST_HEAD(&od->ddev.channels);
 	spin_lock_init(&od->lock);
-- 
2.7.0

[PATCH 2/8] dmaengine: bcm2835: remove unnecessary masking of dma channels

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

The original patch contained 3 dma channels that were masked out.

These - as far as research and discussions show - are a
artefacts remaining from the downstream legacy dma-api.

Right now down-stream still includes a legacy api used only
in a single (downstream only) driver (bcm2708_fb) that requires
2D DMA for speedup (DMA-channel 0).
Formerly the sd-card support driver also was using this legacy
api (DMA-channel 2), but since has been moved over to use
dmaengine directly.

The DMA-channel 3 is already masked out in the devicetree in
the default property "brcm,dma-channel-mask = <0x7f35>;"

So we can remove the whole masking of DMA channels.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index 2d72fe8..e4ca980 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -123,9 +123,6 @@ struct bcm2835_desc {
 #define BCM2835_DMA_DATA_TYPE_S32	4
 #define BCM2835_DMA_DATA_TYPE_S128	16
 
-#define BCM2835_DMA_BULK_MASK	BIT(0)
-#define BCM2835_DMA_FIQ_MASK	(BIT(2) | BIT(3))
-
 /* Valid only for channels 0 - 14, 15 has its own base address */
 #define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
@@ -641,12 +638,6 @@ static int bcm2835_dma_probe(struct platform_device *pdev)
 		goto err_no_dma;
 	}
 
-	/*
-	 * Do not use the FIQ and BULK channels,
-	 * because they are used by the GPU.
-	 */
-	chans_available &= ~(BCM2835_DMA_FIQ_MASK | BCM2835_DMA_BULK_MASK);
-
 	for (i = 0; i < pdev->num_resources; i++) {
 		irq = platform_get_irq(pdev, i);
 		if (irq < 0)
-- 
2.7.0

[PATCH 3/8] dmaengine: bcm2835: add additional defines for DMA-registers

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

Add additional defines describing the DMA registers
as well as adding some more documentation to those registers.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 57 ++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 49 insertions(+), 8 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index e4ca980..a1d851a 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -97,26 +97,67 @@ struct bcm2835_desc {
 
 #define BCM2835_DMA_CS		0x00
 #define BCM2835_DMA_ADDR	0x04
+#define BCM2835_DMA_TI		0x08
 #define BCM2835_DMA_SOURCE_AD	0x0c
 #define BCM2835_DMA_DEST_AD	0x10
-#define BCM2835_DMA_NEXTCB	0x1C
+#define BCM2835_DMA_LEN		0x14
+#define BCM2835_DMA_STRIDE	0x18
+#define BCM2835_DMA_NEXTCB	0x1c
+#define BCM2835_DMA_DEBUG	0x20
 
 /* DMA CS Control and Status bits */
-#define BCM2835_DMA_ACTIVE	BIT(0)
-#define BCM2835_DMA_INT	BIT(2)
+#define BCM2835_DMA_ACTIVE	BIT(0)  /* activate the DMA */
+#define BCM2835_DMA_END		BIT(1)  /* current CB has ended */
+#define BCM2835_DMA_INT		BIT(2)  /* interrupt status */
+#define BCM2835_DMA_DREQ	BIT(3)  /* DREQ state */
 #define BCM2835_DMA_ISPAUSED	BIT(4)  /* Pause requested or not active */
 #define BCM2835_DMA_ISHELD	BIT(5)  /* Is held by DREQ flow control */
-#define BCM2835_DMA_ERR	BIT(8)
+#define BCM2835_DMA_WAITING_FOR_WRITES BIT(6) /* waiting for last
+					       * AXI-write to ack
+					       */
+#define BCM2835_DMA_ERR		BIT(8)
+#define BCM2835_DMA_PRIORITY(x) ((x & 15) << 16) /* AXI priority */
+#define BCM2835_DMA_PANIC_PRIORITY(x) ((x & 15) << 20) /* panic priority */
+/* current value of TI.BCM2835_DMA_WAIT_RESP */
+#define BCM2835_DMA_WAIT_FOR_WRITES BIT(28)
+#define BCM2835_DMA_DIS_DEBUG	BIT(29) /* disable debug pause signal */
 #define BCM2835_DMA_ABORT	BIT(30) /* Stop current CB, go to next, WO */
 #define BCM2835_DMA_RESET	BIT(31) /* WO, self clearing */
 
+/* Transfer information bits - also bcm2835_cb.info field */
 #define BCM2835_DMA_INT_EN	BIT(0)
+#define BCM2835_DMA_TDMODE	BIT(1) /* 2D-Mode */
+#define BCM2835_DMA_WAIT_RESP	BIT(3) /* wait for AXI-write to be acked */
 #define BCM2835_DMA_D_INC	BIT(4)
-#define BCM2835_DMA_D_DREQ	BIT(6)
+#define BCM2835_DMA_D_WIDTH	BIT(5) /* 128bit writes if set */
+#define BCM2835_DMA_D_DREQ	BIT(6) /* enable DREQ for destination */
+#define BCM2835_DMA_D_IGNORE	BIT(7) /* ignore destination writes */
 #define BCM2835_DMA_S_INC	BIT(8)
-#define BCM2835_DMA_S_DREQ	BIT(10)
-
-#define BCM2835_DMA_PER_MAP(x)	((x) << 16)
+#define BCM2835_DMA_S_WIDTH	BIT(9) /* 128bit writes if set */
+#define BCM2835_DMA_S_DREQ	BIT(10) /* enable SREQ for source */
+#define BCM2835_DMA_S_IGNORE	BIT(11) /* ignore source reads - read 0 */
+#define BCM2835_DMA_BURST_LENGTH(x) ((x & 15) << 12)
+#define BCM2835_DMA_PER_MAP(x)	((x & 31) << 16) /* REQ source */
+#define BCM2835_DMA_WAIT(x)	((x & 31) << 21) /* add DMA-wait cycles */
+#define BCM2835_DMA_NO_WIDE_BURSTS BIT(26) /* no 2 beat write bursts */
+
+/* debug register bits */
+#define BCM2835_DMA_DEBUG_LAST_NOT_SET_ERR	BIT(0)
+#define BCM2835_DMA_DEBUG_FIFO_ERR		BIT(1)
+#define BCM2835_DMA_DEBUG_READ_ERR		BIT(2)
+#define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_SHIFT 4
+#define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_BITS 4
+#define BCM2835_DMA_DEBUG_ID_SHIFT		16
+#define BCM2835_DMA_DEBUG_ID_BITS		9
+#define BCM2835_DMA_DEBUG_STATE_SHIFT		16
+#define BCM2835_DMA_DEBUG_STATE_BITS		9
+#define BCM2835_DMA_DEBUG_VERSION_SHIFT		25
+#define BCM2835_DMA_DEBUG_VERSION_BITS		3
+#define BCM2835_DMA_DEBUG_LITE			BIT(28)
+
+/* shared registers for all dma channels */
+#define BCM2835_DMA_INT_STATUS         0xfe0
+#define BCM2835_DMA_ENABLE             0xff0
 
 #define BCM2835_DMA_DATA_TYPE_S8	1
 #define BCM2835_DMA_DATA_TYPE_S16	2
-- 
2.7.0

[PATCH 4/8] dmaengine: bcm2835: move cyclic member from bcm2835_chan into bcm2835_desc

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

In preparation to consolidating code we move the cyclic member
into the bcm_2835_desc structure.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index a1d851a..b3bc382 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -73,7 +73,6 @@ struct bcm2835_chan {
 	struct list_head node;
 
 	struct dma_slave_config	cfg;
-	bool cyclic;
 	unsigned int dreq;
 
 	int ch;
@@ -93,6 +92,8 @@ struct bcm2835_desc {
 
 	unsigned int frames;
 	size_t size;
+
+	bool cyclic;
 };
 
 #define BCM2835_DMA_CS		0x00
@@ -377,8 +378,6 @@ static void bcm2835_dma_issue_pending(struct dma_chan *chan)
 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
 	unsigned long flags;
 
-	c->cyclic = true; /* Nothing else is implemented */
-
 	spin_lock_irqsave(&c->vc.lock, flags);
 	if (vchan_issue_pending(&c->vc) && !c->desc)
 		bcm2835_dma_start_desc(c);
@@ -432,6 +431,7 @@ static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 	d->c = c;
 	d->dir = direction;
 	d->frames = buf_len / period_len;
+	d->cyclic = true;
 
 	d->cb_list = kcalloc(d->frames, sizeof(*d->cb_list), GFP_KERNEL);
 	if (!d->cb_list) {
-- 
2.7.0

[PATCH 5/8] dmaengine: bcm2835: move controlblock chain generation into separate method

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

In preparation of adding slave_sg functionality this patch moves the
generation/allocation of bcm2835_desc and the building of
the corresponding DMA-control-block chain from bcm2835_dma_prep_dma_cyclic
into the newly created method bcm2835_dma_create_cb_chain.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 294 +++++++++++++++++++++++++++++++---------------
 1 file changed, 198 insertions(+), 96 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index b3bc382..4db0e23 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -88,12 +88,12 @@ struct bcm2835_desc {
 	struct virt_dma_desc vd;
 	enum dma_transfer_direction dir;
 
-	struct bcm2835_cb_entry *cb_list;
-
 	unsigned int frames;
 	size_t size;
 
 	bool cyclic;
+
+	struct bcm2835_cb_entry cb_list[];
 };
 
 #define BCM2835_DMA_CS		0x00
@@ -169,6 +169,13 @@ struct bcm2835_desc {
 #define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
 
+/* how many frames of max_len size do we need to transfer len bytes */
+static inline size_t bcm2835_dma_frames_for_length(size_t len,
+						   size_t max_len)
+{
+	return DIV_ROUND_UP(len, max_len);
+}
+
 static inline struct bcm2835_dmadev *to_bcm2835_dma_dev(struct dma_device *d)
 {
 	return container_of(d, struct bcm2835_dmadev, ddev);
@@ -185,19 +192,161 @@ static inline struct bcm2835_desc *to_bcm2835_dma_desc(
 	return container_of(t, struct bcm2835_desc, vd.tx);
 }
 
-static void bcm2835_dma_desc_free(struct virt_dma_desc *vd)
+static void bcm2835_dma_free_cb_chain(struct bcm2835_desc *desc)
 {
-	struct bcm2835_desc *desc = container_of(vd, struct bcm2835_desc, vd);
-	int i;
+	size_t i;
 
 	for (i = 0; i < desc->frames; i++)
 		dma_pool_free(desc->c->cb_pool, desc->cb_list[i].cb,
 			      desc->cb_list[i].paddr);
 
-	kfree(desc->cb_list);
 	kfree(desc);
 }
 
+static void bcm2835_dma_desc_free(struct virt_dma_desc *vd)
+{
+	bcm2835_dma_free_cb_chain(
+		container_of(vd, struct bcm2835_desc, vd));
+}
+
+static void bcm2835_dma_create_cb_set_length(
+	struct bcm2835_chan *chan,
+	struct bcm2835_dma_cb *control_block,
+	size_t len,
+	size_t period_len,
+	size_t *total_len,
+	u32 finalextrainfo)
+{
+	/* set the length */
+	control_block->length = len;
+
+	/* finished if we have no period_length */
+	if (!period_len)
+		return;
+
+	/*
+	 * period_len means: that we need to generate
+	 * transfers that are terminating at every
+	 * multiple of period_len - this is typically
+	 * used to set the interrupt flag in info
+	 * which is required during cyclic transfers
+	 */
+
+	/* have we filled in period_length yet? */
+	if (*total_len + control_block->length < period_len)
+		return;
+
+	/* calculate the length that remains to reach period_length */
+	control_block->length = period_len - *total_len;
+
+	/* reset total_length for next period */
+	*total_len = 0;
+
+	/* add extrainfo bits in info */
+	control_block->info |= finalextrainfo;
+}
+
+/**
+ * bcm2835_dma_create_cb_chain - create a control block and fills data in
+ *
+ * @chan:           the @dma_chan for which we run this
+ * @direction:      the direction in which we transfer
+ * @cyclic:         it is a cyclic transfer
+ * @info:           the default info bits to apply per controlblock
+ * @frames:         number of controlblocks to allocate
+ * @src:            the src address to assign (if the S_INC bit is set
+ *                  in @info, then it gets incremented)
+ * @dst:            the dst address to assign (if the D_INC bit is set
+ *                  in @info, then it gets incremented)
+ * @buf_len:        the full buffer length (may also be 0)
+ * @period_len:     the period length when to apply @finalextrainfo
+ *                  in addition to the last transfer
+ *                  this will also break some control-blocks early
+ * @finalextrainfo: additional bits in last controlblock
+ *                  (or when period_len is reached in case of cyclic)
+ * @gfp:            the GFP flag to use for allocation
+ */
+static struct bcm2835_desc *bcm2835_dma_create_cb_chain(
+	struct dma_chan *chan, enum dma_transfer_direction direction,
+	bool cyclic, u32 info, u32 finalextrainfo, size_t frames,
+	dma_addr_t src, dma_addr_t dst, size_t buf_len,
+	size_t period_len, gfp_t gfp)
+{
+	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
+	size_t len = buf_len, total_len;
+	size_t frame;
+	struct bcm2835_desc *d;
+	struct bcm2835_cb_entry *cb_entry;
+	struct bcm2835_dma_cb *control_block;
+
+	/* allocate and setup the descriptor. */
+	d = kzalloc(sizeof(*d) + frames * sizeof(struct bcm2835_cb_entry),
+		    gfp);
+	if (!d)
+		return NULL;
+
+	d->c = c;
+	d->dir = direction;
+	d->cyclic = cyclic;
+
+	/*
+	 * Iterate over all frames, create a control block
+	 * for each frame and link them together.
+	 */
+	for (frame = 0, total_len = 0; frame < frames; d->frames++, frame++) {
+		cb_entry = &d->cb_list[frame];
+		cb_entry->cb = dma_pool_alloc(c->cb_pool, gfp,
+					      &cb_entry->paddr);
+		if (!cb_entry->cb)
+			goto error_cb;
+
+		/* fill in the control block */
+		control_block = cb_entry->cb;
+		control_block->info = info;
+		control_block->src = src;
+		control_block->dst = dst;
+		control_block->stride = 0;
+		control_block->next = 0;
+		/* set up length in control_block if requested */
+		if (buf_len) {
+			/* calculate length honoring period_length */
+			bcm2835_dma_create_cb_set_length(
+				c, control_block,
+				len, period_len, &total_len,
+				cyclic ? finalextrainfo : 0);
+
+			/* calculate new remaining length */
+			len -= control_block->length;
+		}
+
+		/* link this the last controlblock */
+		if (frame)
+			d->cb_list[frame - 1].cb->next = cb_entry->paddr;
+
+		/* update src and dst and length */
+		if (src && (info & BCM2835_DMA_S_INC))
+			src += control_block->length;
+		if (dst && (info & BCM2835_DMA_D_INC))
+			dst += control_block->length;
+
+		/* Length of total transfer */
+		d->size += control_block->length;
+	}
+
+	/* the last frame requires extra flags */
+	d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
+
+	/* detect a size missmatch */
+	if (buf_len && (d->size != buf_len))
+		goto error_cb;
+
+	return d;
+error_cb:
+	bcm2835_dma_free_cb_chain(d);
+
+	return NULL;
+}
+
 static int bcm2835_dma_abort(void __iomem *chan_base)
 {
 	unsigned long cs;
@@ -391,12 +540,11 @@ static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 	unsigned long flags)
 {
 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
-	enum dma_slave_buswidth dev_width;
 	struct bcm2835_desc *d;
-	dma_addr_t dev_addr;
-	unsigned int es, sync_type;
-	unsigned int frame;
-	int i;
+	dma_addr_t src, dst;
+	u32 info = BCM2835_DMA_WAIT_RESP;
+	u32 extra = BCM2835_DMA_INT_EN;
+	size_t frames;
 
 	/* Grab configuration */
 	if (!is_slave_direction(direction)) {
@@ -404,104 +552,58 @@ static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 		return NULL;
 	}
 
-	if (direction == DMA_DEV_TO_MEM) {
-		dev_addr = c->cfg.src_addr;
-		dev_width = c->cfg.src_addr_width;
-		sync_type = BCM2835_DMA_S_DREQ;
-	} else {
-		dev_addr = c->cfg.dst_addr;
-		dev_width = c->cfg.dst_addr_width;
-		sync_type = BCM2835_DMA_D_DREQ;
-	}
-
-	/* Bus width translates to the element size (ES) */
-	switch (dev_width) {
-	case DMA_SLAVE_BUSWIDTH_4_BYTES:
-		es = BCM2835_DMA_DATA_TYPE_S32;
-		break;
-	default:
+	if (!buf_len) {
+		dev_err(chan->device->dev,
+			"%s: bad buffer length (= 0)\n", __func__);
 		return NULL;
 	}
 
-	/* Now allocate and setup the descriptor. */
-	d = kzalloc(sizeof(*d), GFP_NOWAIT);
-	if (!d)
-		return NULL;
+	/*
+	 * warn if buf_len is not a multiple of period_len - this may leed
+	 * to unexpected latencies for interrupts and thus audiable clicks
+	 */
+	if (buf_len % period_len)
+		dev_warn_once(chan->device->dev,
+			      "%s: buffer_length (%zd) is not a multiple of period_len (%zd)\n",
+			      __func__, buf_len, period_len);
 
-	d->c = c;
-	d->dir = direction;
-	d->frames = buf_len / period_len;
-	d->cyclic = true;
+	/* Setup DREQ channel */
+	if (c->dreq != 0)
+		info |= BCM2835_DMA_PER_MAP(c->dreq);
 
-	d->cb_list = kcalloc(d->frames, sizeof(*d->cb_list), GFP_KERNEL);
-	if (!d->cb_list) {
-		kfree(d);
-		return NULL;
+	if (direction == DMA_DEV_TO_MEM) {
+		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+			return NULL;
+		src = c->cfg.src_addr;
+		dst = buf_addr;
+		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
+	} else {
+		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+			return NULL;
+		dst = c->cfg.dst_addr;
+		src = buf_addr;
+		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
 	}
-	/* Allocate memory for control blocks */
-	for (i = 0; i < d->frames; i++) {
-		struct bcm2835_cb_entry *cb_entry = &d->cb_list[i];
 
-		cb_entry->cb = dma_pool_zalloc(c->cb_pool, GFP_ATOMIC,
-					       &cb_entry->paddr);
-		if (!cb_entry->cb)
-			goto error_cb;
-	}
+	/* calculate number of frames */
+	frames = DIV_ROUND_UP(buf_len, period_len);
 
 	/*
-	 * Iterate over all frames, create a control block
-	 * for each frame and link them together.
+	 * allocate the CB chain
+	 * note that we need to use GFP_NOWAIT, as the ALSA i2s dmaengine
+	 * implementation calls prep_dma_cyclic with interrupts disabled.
 	 */
-	for (frame = 0; frame < d->frames; frame++) {
-		struct bcm2835_dma_cb *control_block = d->cb_list[frame].cb;
-
-		/* Setup adresses */
-		if (d->dir == DMA_DEV_TO_MEM) {
-			control_block->info = BCM2835_DMA_D_INC;
-			control_block->src = dev_addr;
-			control_block->dst = buf_addr + frame * period_len;
-		} else {
-			control_block->info = BCM2835_DMA_S_INC;
-			control_block->src = buf_addr + frame * period_len;
-			control_block->dst = dev_addr;
-		}
-
-		/* Enable interrupt */
-		control_block->info |= BCM2835_DMA_INT_EN;
-
-		/* Setup synchronization */
-		if (sync_type != 0)
-			control_block->info |= sync_type;
-
-		/* Setup DREQ channel */
-		if (c->dreq != 0)
-			control_block->info |=
-				BCM2835_DMA_PER_MAP(c->dreq);
-
-		/* Length of a frame */
-		control_block->length = period_len;
-		d->size += control_block->length;
+	d = bcm2835_dma_create_cb_chain(chan, direction, true,
+					info, extra,
+					frames, src, dst, buf_len,
+					period_len, GFP_NOWAIT);
+	if (!d)
+		return NULL;
 
-		/*
-		 * Next block is the next frame.
-		 * This DMA engine driver currently only supports cyclic DMA.
-		 * Therefore, wrap around at number of frames.
-		 */
-		control_block->next = d->cb_list[((frame + 1) % d->frames)].paddr;
-	}
+	/* wrap around into a loop */
+	d->cb_list[d->frames - 1].cb->next = d->cb_list[0].paddr;
 
 	return vchan_tx_prep(&c->vc, &d->vd, flags);
-error_cb:
-	i--;
-	for (; i >= 0; i--) {
-		struct bcm2835_cb_entry *cb_entry = &d->cb_list[i];
-
-		dma_pool_free(c->cb_pool, cb_entry->cb, cb_entry->paddr);
-	}
-
-	kfree(d->cb_list);
-	kfree(d);
-	return NULL;
 }
 
 static int bcm2835_dma_slave_config(struct dma_chan *chan,
-- 
2.7.0

[PATCH 6/8] dmaengine: bcm2835: limit max length based on channel type

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

The bcm2835 dma system has 2 basic types of dma-channels:
* "normal" channels
* "light" channels

Lite channels are limited in several aspects:
* internal data-structure is 128 bit (not 256)
* does not support BCM2835_DMA_TDMODE (2D)
* DMA length register is limited to 16 bit.
  so 0-65535 (not 0-65536 as mentioned in the official datasheet)
* BCM2835_DMA_S/D_IGNORE are not supported

The detection of the type of mode is implemented by looking at
the LITE bit in the DEBUG register for each channel.
This allows automatic detection.

Based on this the maximum block size is set to (64K - 4) or to 1G
and this limit is honored during generation of control block
chains. The effect is that when a LITE channel is used more
control blocks are used to do the same transfer (compared
to a normal channel).

As there are several sources/target DREQS that are 32 bit wide
we need to have the transfer to be a multiple of 4 as this would
break the transfer otherwise.

This is why the limit of (64K - 4) was chosen over the
alternative of (64K - 4K).

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 29 ++++++++++++++++++++++++++---
 1 file changed, 26 insertions(+), 3 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index 4db0e23..59c5ef3 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -81,6 +81,8 @@ struct bcm2835_chan {
 
 	void __iomem *chan_base;
 	int irq_number;
+
+	bool is_lite_channel;
 };
 
 struct bcm2835_desc {
@@ -169,6 +171,16 @@ struct bcm2835_desc {
 #define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
 
+/* the max dma length for different channels */
+#define MAX_DMA_LEN SZ_1G
+#define MAX_LITE_DMA_LEN (SZ_64K - 4)
+
+static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
+{
+	/* lite and normal channels have different max frame length */
+	return c->is_lite_channel ? MAX_LITE_DMA_LEN : MAX_DMA_LEN;
+}
+
 /* how many frames of max_len size do we need to transfer len bytes */
 static inline size_t bcm2835_dma_frames_for_length(size_t len,
 						   size_t max_len)
@@ -217,8 +229,10 @@ static void bcm2835_dma_create_cb_set_length(
 	size_t *total_len,
 	u32 finalextrainfo)
 {
-	/* set the length */
-	control_block->length = len;
+	size_t max_len = bcm2835_dma_max_frame_length(chan);
+
+	/* set the length taking lite-channel limitations into account */
+	control_block->length = min_t(u32, len, max_len);
 
 	/* finished if we have no period_length */
 	if (!period_len)
@@ -544,6 +558,7 @@ static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 	dma_addr_t src, dst;
 	u32 info = BCM2835_DMA_WAIT_RESP;
 	u32 extra = BCM2835_DMA_INT_EN;
+	size_t max_len = bcm2835_dma_max_frame_length(c);
 	size_t frames;
 
 	/* Grab configuration */
@@ -586,7 +601,10 @@ static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 	}
 
 	/* calculate number of frames */
-	frames = DIV_ROUND_UP(buf_len, period_len);
+	frames = /* number of periods */
+		 DIV_ROUND_UP(buf_len, period_len) *
+		 /* number of frames per period */
+		 bcm2835_dma_frames_for_length(period_len, max_len);
 
 	/*
 	 * allocate the CB chain
@@ -685,6 +703,11 @@ static int bcm2835_dma_chan_init(struct bcm2835_dmadev *d, int chan_id, int irq)
 	c->ch = chan_id;
 	c->irq_number = irq;
 
+	/* check in DEBUG register if this is a LITE channel */
+	if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
+		BCM2835_DMA_DEBUG_LITE)
+		c->is_lite_channel = true;
+
 	return 0;
 }
 
-- 
2.7.0

[PATCH 7/8] dmaengine: bcm2835: add slave_sg support to bcm2835-dma

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

Add slave_sg support to bcm2835-dma using shared allocation
code for bcm2835_desc and DMA-control blocks already used by
dma_cyclic.

Note that bcm2835_dma_callback had to get modified to support
both modes of operation (cyclic and non-cyclic).

Tested using:
* Hifiberry I2S card (using cyclic DMA)
* fb_st7735r SPI-framebuffer (using slave_sg DMA via spi-bcm2835)
playing BigBuckBunny for audio and video.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 113 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 108 insertions(+), 5 deletions(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index 59c5ef3..b46b12f 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -260,6 +260,23 @@ static void bcm2835_dma_create_cb_set_length(
 	control_block->info |= finalextrainfo;
 }
 
+static inline size_t bcm2835_dma_count_frames_for_sg(
+	struct bcm2835_chan *c,
+	struct scatterlist *sgl,
+	unsigned int sg_len)
+{
+	size_t frames = 0;
+	struct scatterlist *sgent;
+	unsigned int i;
+	size_t plength = bcm2835_dma_max_frame_length(c);
+
+	for_each_sg(sgl, sgent, sg_len, i)
+		frames += bcm2835_dma_frames_for_length(
+			sg_dma_len(sgent), plength);
+
+	return frames;
+}
+
 /**
  * bcm2835_dma_create_cb_chain - create a control block and fills data in
  *
@@ -361,6 +378,32 @@ error_cb:
 	return NULL;
 }
 
+static void bcm2835_dma_fill_cb_chain_with_sg(
+	struct dma_chan *chan,
+	enum dma_transfer_direction direction,
+	struct bcm2835_cb_entry *cb,
+	struct scatterlist *sgl,
+	unsigned int sg_len)
+{
+	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
+	size_t max_len = bcm2835_dma_max_frame_length(c);
+	unsigned int i, len;
+	dma_addr_t addr;
+	struct scatterlist *sgent;
+
+	for_each_sg(sgl, sgent, sg_len, i) {
+		for (addr = sg_dma_address(sgent), len = sg_dma_len(sgent);
+		     len > 0;
+		     addr += cb->cb->length, len -= cb->cb->length, cb++) {
+			if (direction == DMA_DEV_TO_MEM)
+				cb->cb->dst = addr;
+			else
+				cb->cb->src = addr;
+			cb->cb->length = min(len, max_len);
+		}
+	}
+}
+
 static int bcm2835_dma_abort(void __iomem *chan_base)
 {
 	unsigned long cs;
@@ -428,13 +471,19 @@ static irqreturn_t bcm2835_dma_callback(int irq, void *data)
 	d = c->desc;
 
 	if (d) {
-		/* TODO Only works for cyclic DMA */
-		vchan_cyclic_callback(&d->vd);
+		if (d->cyclic) {
+			/* call the cyclic callback */
+			vchan_cyclic_callback(&d->vd);
+
+			/* Keep the DMA engine running */
+			writel(BCM2835_DMA_ACTIVE,
+			       c->chan_base + BCM2835_DMA_CS);
+		} else {
+			vchan_cookie_complete(&c->desc->vd);
+			bcm2835_dma_start_desc(c);
+		}
 	}
 
-	/* Keep the DMA engine running */
-	writel(BCM2835_DMA_ACTIVE, c->chan_base + BCM2835_DMA_CS);
-
 	spin_unlock_irqrestore(&c->vc.lock, flags);
 
 	return IRQ_HANDLED;
@@ -548,6 +597,58 @@ static void bcm2835_dma_issue_pending(struct dma_chan *chan)
 	spin_unlock_irqrestore(&c->vc.lock, flags);
 }
 
+static struct dma_async_tx_descriptor *bcm2835_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 bcm2835_chan *c = to_bcm2835_dma_chan(chan);
+	struct bcm2835_desc *d;
+	dma_addr_t src = 0, dst = 0;
+	u32 info = BCM2835_DMA_WAIT_RESP;
+	u32 extra = BCM2835_DMA_INT_EN;
+	size_t frames;
+
+	if (!is_slave_direction(direction)) {
+		dev_err(chan->device->dev,
+			"%s: bad direction?\n", __func__);
+		return NULL;
+	}
+
+	if (c->dreq != 0)
+		info |= BCM2835_DMA_PER_MAP(c->dreq);
+
+	if (direction == DMA_DEV_TO_MEM) {
+		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+			return NULL;
+		src = c->cfg.src_addr;
+		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
+	} else {
+		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+			return NULL;
+		dst = c->cfg.dst_addr;
+		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
+	}
+
+	/* count frames in sg list */
+	frames = bcm2835_dma_count_frames_for_sg(c, sgl, sg_len);
+
+	/* allocate the CB chain */
+	d = bcm2835_dma_create_cb_chain(chan, direction, false,
+					info, extra,
+					frames, src, dst, 0, 0,
+					GFP_KERNEL);
+	if (!d)
+		return NULL;
+
+	/* fill in frames with scatterlist pointers */
+	bcm2835_dma_fill_cb_chain_with_sg(chan, direction, d->cb_list,
+					  sgl, sg_len);
+
+	return vchan_tx_prep(&c->vc, &d->vd, flags);
+}
+
 static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
 	struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
 	size_t period_len, enum dma_transfer_direction direction,
@@ -778,11 +879,13 @@ static int bcm2835_dma_probe(struct platform_device *pdev)
 	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
 	dma_cap_set(DMA_PRIVATE, od->ddev.cap_mask);
 	dma_cap_set(DMA_CYCLIC, od->ddev.cap_mask);
+	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
 	od->ddev.device_alloc_chan_resources = bcm2835_dma_alloc_chan_resources;
 	od->ddev.device_free_chan_resources = bcm2835_dma_free_chan_resources;
 	od->ddev.device_tx_status = bcm2835_dma_tx_status;
 	od->ddev.device_issue_pending = bcm2835_dma_issue_pending;
 	od->ddev.device_prep_dma_cyclic = bcm2835_dma_prep_dma_cyclic;
+	od->ddev.device_prep_slave_sg = bcm2835_dma_prep_slave_sg;
 	od->ddev.device_config = bcm2835_dma_slave_config;
 	od->ddev.device_terminate_all = bcm2835_dma_terminate_all;
 	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
-- 
2.7.0

[PATCH 8/8] dmaengine: bcm2835: add dma_memcopy support to bcm2835-dma

[Eric Anholt]

From: Martin Sperl <kernel@martin.sperl.org>

Also added check for an error condition in bcm2835_dma_create_cb_chain
that showed up during development of this patch.

Tested using dmatest for all enabled channels.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/dma/bcm2835-dma.c | 36 +++++++++++++++++++++++++++++++++++-
 1 file changed, 35 insertions(+), 1 deletion(-)

diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
index b46b12f..cc771cd 100644
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -310,6 +310,9 @@ static struct bcm2835_desc *bcm2835_dma_create_cb_chain(
 	struct bcm2835_cb_entry *cb_entry;
 	struct bcm2835_dma_cb *control_block;
 
+	if (!frames)
+		return NULL;
+
 	/* allocate and setup the descriptor. */
 	d = kzalloc(sizeof(*d) + frames * sizeof(struct bcm2835_cb_entry),
 		    gfp);
@@ -597,6 +600,34 @@ static void bcm2835_dma_issue_pending(struct dma_chan *chan)
 	spin_unlock_irqrestore(&c->vc.lock, flags);
 }
 
+struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_memcpy(
+	struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
+	size_t len, unsigned long flags)
+{
+	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
+	struct bcm2835_desc *d;
+	u32 info = BCM2835_DMA_D_INC | BCM2835_DMA_S_INC;
+	u32 extra = BCM2835_DMA_INT_EN | BCM2835_DMA_WAIT_RESP;
+	size_t max_len = bcm2835_dma_max_frame_length(c);
+	size_t frames;
+
+	/* if src, dst or len is not given return with an error */
+	if (!src || !dst || !len)
+		return NULL;
+
+	/* calculate number of frames */
+	frames = bcm2835_dma_frames_for_length(len, max_len);
+
+	/* allocate the CB chain - this also fills in the pointers */
+	d = bcm2835_dma_create_cb_chain(chan, DMA_MEM_TO_MEM, false,
+					info, extra, frames,
+					src, dst, len, 0, GFP_KERNEL);
+	if (!d)
+		return NULL;
+
+	return vchan_tx_prep(&c->vc, &d->vd, flags);
+}
+
 static struct dma_async_tx_descriptor *bcm2835_dma_prep_slave_sg(
 	struct dma_chan *chan,
 	struct scatterlist *sgl, unsigned int sg_len,
@@ -880,17 +911,20 @@ static int bcm2835_dma_probe(struct platform_device *pdev)
 	dma_cap_set(DMA_PRIVATE, od->ddev.cap_mask);
 	dma_cap_set(DMA_CYCLIC, od->ddev.cap_mask);
 	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
+	dma_cap_set(DMA_MEMCPY, od->ddev.cap_mask);
 	od->ddev.device_alloc_chan_resources = bcm2835_dma_alloc_chan_resources;
 	od->ddev.device_free_chan_resources = bcm2835_dma_free_chan_resources;
 	od->ddev.device_tx_status = bcm2835_dma_tx_status;
 	od->ddev.device_issue_pending = bcm2835_dma_issue_pending;
 	od->ddev.device_prep_dma_cyclic = bcm2835_dma_prep_dma_cyclic;
 	od->ddev.device_prep_slave_sg = bcm2835_dma_prep_slave_sg;
+	od->ddev.device_prep_dma_memcpy = bcm2835_dma_prep_dma_memcpy;
 	od->ddev.device_config = bcm2835_dma_slave_config;
 	od->ddev.device_terminate_all = bcm2835_dma_terminate_all;
 	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
 	od->ddev.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
-	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
+	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV) |
+			      BIT(DMA_MEM_TO_MEM);
 	od->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
 	od->ddev.dev = &pdev->dev;
 	INIT_LIST_HEAD(&od->ddev.channels);
-- 
2.7.0

[PATCH 0/8 v4] bcm2835 DMA slave support

[Martin Sperl]

On 16.03.2016 20:24, Eric Anholt wrote:
> Here's the series for DMA slave and memcpy support for 2835, with the
> DT changes to enable the remaining channels dropped out while that
> goes through review.  I had to do some minor conflict resolution, but
> it was pretty mechanical, and I tested again with dmatest on the last
> patch.
I guess I have got a list to implement that without a change to the 
device tree,
besides making use of platform_get_irq_byname and setting up the property:
     interrupt-names = "dma0", "dma1", .. "dma10", "dma_shared", "dma_all";

As the property "interrupt-names" is a standard it should be acceptable.

[PATCH 0/8 v4] bcm2835 DMA slave support

[Martin Sperl]

On 16.03.2016 20:24, Eric Anholt wrote:
> Here's the series for DMA slave and memcpy support for 2835, with the
> DT changes to enable the remaining channels dropped out while that
> goes through review.  I had to do some minor conflict resolution, but
> it was pretty mechanical, and I tested again with dmatest on the last
> patch.
>
> Martin Sperl (8):
>    dmaengine: bcm2835: set residue_granularity field
>    dmaengine: bcm2835: remove unnecessary masking of dma channels
>    dmaengine: bcm2835: add additional defines for DMA-registers
>    dmaengine: bcm2835: move cyclic member from bcm2835_chan into
>      bcm2835_desc
>    dmaengine: bcm2835: move controlblock chain generation into separate
>      method
>    dmaengine: bcm2835: limit max length based on channel type
>    dmaengine: bcm2835: add slave_sg support to bcm2835-dma
>    dmaengine: bcm2835: add dma_memcopy support to bcm2835-dma
>
I have successfully tested this modified patch-series
playing BigBuckBunny on:
* fb-tft device (fb_st7735r) - via spi-bcm2835 using slave_sg dma
* I2S Hifiberry DAC (snd_soc_hifiberry_dac) - via bcm2835-i2s using 
cyclic dma

Required additional patches to make this work
(especially I2S support, which is non-working since ):
* the clock-patchsets:
  * [PATCH 0/6] clk: bcm2835: fixes clk-bcm2835 driver issues
     (most are reviewed by Eric)
  * [PATCH 0/3] reorganize clock initialization and add PCM clock
     (no reviewed/acked so far)
* i2s patchset to enable the use of the clock framework
  * [PATCH V2 0/3] ASOC: bcm2835: move bcm2835-i2s to use clock framework
     (if I remember correctly Mark Brown has merged the driver patches)
* out of tree drivers for Hifiberry DAC
    (I guess I should upstream those...)

Tested-by: Martin Sperl <kernel@martin.sperl.org>

[PATCH 0/8 v4] bcm2835 DMA slave support

[Vinod Koul]

On Wed, Mar 16, 2016 at 12:24:54PM -0700, Eric Anholt wrote:
> Here's the series for DMA slave and memcpy support for 2835, with the
> DT changes to enable the remaining channels dropped out while that
> goes through review.  I had to do some minor conflict resolution, but
> it was pretty mechanical, and I tested again with dmatest on the last
> patch.

Applied, thanks

-- 
~Vinod