From 621fb1606217c3e72feda69255ae6cb6a7ccfec2 Mon Sep 17 00:00:00 2001
From: Phil Elwell <phil@raspberrypi.org>
Date: Thu, 4 Apr 2019 13:33:47 +0100
Subject: [PATCH] bcm2835-dma: Add proper 40-bit DMA support

The 40-bit additions are not fully tested, but it should be
capable of supporting both 40-bit memcpy on BCM2711 and regular
Lite channels on BCM2835.

Signed-off-by: Phil Elwell <phil@raspberrypi.org>
---
 arch/arm/boot/dts/bcm2838.dtsi               |  33 +-
 drivers/dma/bcm2835-dma.c                    | 426 ++++++++++++++-----
 drivers/pci/controller/pcie-brcmstb-bounce.c |  30 +-
 drivers/pci/controller/pcie-brcmstb-bounce.h |  21 +-
 drivers/pci/controller/pcie-brcmstb.c        |  23 +-
 5 files changed, 395 insertions(+), 138 deletions(-)

--- a/arch/arm/boot/dts/bcm2838.dtsi
+++ b/arch/arm/boot/dts/bcm2838.dtsi
@@ -372,6 +372,23 @@
 			};
 		};
 
+		dma40: dma@7e007b00 {
+			compatible = "brcm,bcm2838-dma";
+			reg = <0x0 0x7e007b00 0x400>;
+			interrupts =
+				<GIC_SPI 89 IRQ_TYPE_LEVEL_HIGH>, /* dma4 11 */
+				<GIC_SPI 90 IRQ_TYPE_LEVEL_HIGH>, /* dma4 12 */
+				<GIC_SPI 91 IRQ_TYPE_LEVEL_HIGH>, /* dma4 13 */
+				<GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>; /* dma4 14 */
+			interrupt-names = "dma11",
+				"dma12",
+				"dma13",
+				"dma14";
+			#dma-cells = <1>;
+			brcm,dma-channel-mask = <0x7000>;
+		};
+		/* DMA4 - 40 bit DMA engines */
+
 		xhci: xhci@7e9c0000 {
 			compatible = "generic-xhci";
 			status = "disabled";
@@ -689,6 +706,7 @@
 };
 
 &dma {
+	reg = <0x7e007000 0xb00>;
 	interrupts = <GIC_SPI 80 IRQ_TYPE_LEVEL_HIGH>,
 		<GIC_SPI 81 IRQ_TYPE_LEVEL_HIGH>,
 		<GIC_SPI 82 IRQ_TYPE_LEVEL_HIGH>,
@@ -699,12 +717,7 @@
 		<GIC_SPI 87 IRQ_TYPE_LEVEL_HIGH>, /* dmalite  7 */
 		<GIC_SPI 87 IRQ_TYPE_LEVEL_HIGH>, /* dmalite  8 */
 		<GIC_SPI 88 IRQ_TYPE_LEVEL_HIGH>, /* dmalite  9 */
-		<GIC_SPI 88 IRQ_TYPE_LEVEL_HIGH>, /* dmalite 10 */
-		/* DMA4 - 40 bit DMA engines */
-		<GIC_SPI 89 IRQ_TYPE_LEVEL_HIGH>, /* dma4    11 */
-		<GIC_SPI 90 IRQ_TYPE_LEVEL_HIGH>, /* dma4    12 */
-		<GIC_SPI 91 IRQ_TYPE_LEVEL_HIGH>, /* dma4    13 */
-		<GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>; /* dma4    14 */
+		<GIC_SPI 88 IRQ_TYPE_LEVEL_HIGH>; /* dmalite 10 */
 	interrupt-names = "dma0",
 			  "dma1",
 			  "dma2",
@@ -715,10 +728,6 @@
 			  "dma7",
 			  "dma8",
 			  "dma9",
-			  "dma10",
-			  "dma11",
-			  "dma12",
-			  "dma13",
-			  "dma14";
-	brcm,dma-channel-mask = <0x7ef5>;
+			  "dma10";
+	brcm,dma-channel-mask = <0x01f5>;
 };
--- a/drivers/dma/bcm2835-dma.c
+++ b/drivers/dma/bcm2835-dma.c
@@ -50,12 +50,18 @@
 #define BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED 14
 #define BCM2835_DMA_CHAN_NAME_SIZE 8
 #define BCM2835_DMA_BULK_MASK  BIT(0)
+#define BCM2838_DMA_MEMCPY_CHAN 14
+
+struct bcm2835_dma_cfg_data {
+	u32	chan_40bit_mask;
+};
 
 struct bcm2835_dmadev {
 	struct dma_device ddev;
 	spinlock_t lock;
 	void __iomem *base;
 	struct device_dma_parameters dma_parms;
+	const struct bcm2835_dma_cfg_data *cfg_data;
 };
 
 struct bcm2835_dma_cb {
@@ -100,6 +106,7 @@ struct bcm2835_chan {
 	unsigned int irq_flags;
 
 	bool is_lite_channel;
+	bool is_40bit_channel;
 };
 
 struct bcm2835_desc {
@@ -189,7 +196,8 @@ struct bcm2835_desc {
 #define BCM2835_DMA_DATA_TYPE_S128	16
 
 /* Valid only for channels 0 - 14, 15 has its own base address */
-#define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
+#define BCM2835_DMA_CHAN_SIZE	0x100
+#define BCM2835_DMA_CHAN(n)	((n) * BCM2835_DMA_CHAN_SIZE) /* Base address */
 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
 
 /* the max dma length for different channels */
@@ -200,7 +208,7 @@ struct bcm2835_desc {
 #define BCM2838_DMA40_CS	0x00
 #define BCM2838_DMA40_CB	0x04
 #define BCM2838_DMA40_DEBUG	0x0c
-#define BCM2858_DMA40_TI	0x10
+#define BCM2838_DMA40_TI	0x10
 #define BCM2838_DMA40_SRC	0x14
 #define BCM2838_DMA40_SRCI	0x18
 #define BCM2838_DMA40_DEST	0x1c
@@ -209,32 +217,97 @@ struct bcm2835_desc {
 #define BCM2838_DMA40_NEXT_CB	0x28
 #define BCM2838_DMA40_DEBUG2	0x2c
 
-#define BCM2838_DMA40_CS_ACTIVE	BIT(0)
-#define BCM2838_DMA40_CS_END	BIT(1)
+#define BCM2838_DMA40_ACTIVE		BIT(0)
+#define BCM2838_DMA40_END		BIT(1)
+#define BCM2838_DMA40_INT		BIT(2)
+#define BCM2838_DMA40_DREQ		BIT(3)  /* DREQ state */
+#define BCM2838_DMA40_RD_PAUSED		BIT(4)  /* Reading is paused */
+#define BCM2838_DMA40_WR_PAUSED		BIT(5)  /* Writing is paused */
+#define BCM2838_DMA40_DREQ_PAUSED	BIT(6)  /* Is paused by DREQ flow control */
+#define BCM2838_DMA40_WAITING_FOR_WRITES BIT(7)  /* Waiting for last write */
+#define BCM2838_DMA40_ERR		BIT(10)
+#define BCM2838_DMA40_QOS(x)		(((x) & 0x1f) << 16)
+#define BCM2838_DMA40_PANIC_QOS(x)	(((x) & 0x1f) << 20)
+#define BCM2838_DMA40_WAIT_FOR_WRITES	BIT(28)
+#define BCM2838_DMA40_DISDEBUG		BIT(29)
+#define BCM2838_DMA40_ABORT		BIT(30)
+#define BCM2838_DMA40_HALT		BIT(31)
+#define BCM2838_DMA40_CS_FLAGS(x) (x & (BCM2838_DMA40_QOS(15) | \
+					BCM2838_DMA40_PANIC_QOS(15) | \
+					BCM2838_DMA40_WAIT_FOR_WRITES |	\
+					BCM2838_DMA40_DISDEBUG))
+
+/* Transfer information bits */
+#define BCM2838_DMA40_INTEN		BIT(0)
+#define BCM2838_DMA40_TDMODE		BIT(1) /* 2D-Mode */
+#define BCM2838_DMA40_WAIT_RESP		BIT(2) /* wait for AXI write to be acked */
+#define BCM2838_DMA40_WAIT_RD_RESP	BIT(3) /* wait for AXI read to complete */
+#define BCM2838_DMA40_PER_MAP(x)	((x & 31) << 9) /* REQ source */
+#define BCM2838_DMA40_S_DREQ		BIT(14) /* enable SREQ for source */
+#define BCM2838_DMA40_D_DREQ		BIT(15) /* enable DREQ for destination */
+#define BCM2838_DMA40_S_WAIT(x)		((x & 0xff) << 16) /* add DMA read-wait cycles */
+#define BCM2838_DMA40_D_WAIT(x)		((x & 0xff) << 24) /* add DMA write-wait cycles */
 
-#define BCM2838_DMA40_CS_QOS(x)	(((x) & 0x1f) << 16)
-#define BCM2838_DMA40_CS_PANIC_QOS(x)	(((x) & 0x1f) << 20)
-#define BCM2838_DMA40_CS_WRITE_WAIT	BIT(28)
+/* debug register bits */
+#define BCM2838_DMA40_DEBUG_WRITE_ERR		BIT(0)
+#define BCM2838_DMA40_DEBUG_FIFO_ERR		BIT(1)
+#define BCM2838_DMA40_DEBUG_READ_ERR		BIT(2)
+#define BCM2838_DMA40_DEBUG_READ_CB_ERR		BIT(3)
+#define BCM2838_DMA40_DEBUG_IN_ON_ERR		BIT(8)
+#define BCM2838_DMA40_DEBUG_ABORT_ON_ERR	BIT(9)
+#define BCM2838_DMA40_DEBUG_HALT_ON_ERR		BIT(10)
+#define BCM2838_DMA40_DEBUG_DISABLE_CLK_GATE	BIT(11)
+#define BCM2838_DMA40_DEBUG_RSTATE_SHIFT	14
+#define BCM2838_DMA40_DEBUG_RSTATE_BITS		4
+#define BCM2838_DMA40_DEBUG_WSTATE_SHIFT	18
+#define BCM2838_DMA40_DEBUG_WSTATE_BITS		4
+#define BCM2838_DMA40_DEBUG_RESET		BIT(23)
+#define BCM2838_DMA40_DEBUG_ID_SHIFT		24
+#define BCM2838_DMA40_DEBUG_ID_BITS		4
+#define BCM2838_DMA40_DEBUG_VERSION_SHIFT	28
+#define BCM2838_DMA40_DEBUG_VERSION_BITS	4
+
+/* Valid only for channels 0 - 3 (11 - 14) */
+#define BCM2838_DMA40_CHAN(n)	(((n) + 11) << 8) /* Base address */
+#define BCM2838_DMA40_CHANIO(base, n) ((base) + BCM2838_DMA_CHAN(n))
 
-#define BCM2838_DMA40_BURST_LEN(x)	((((x) - 1) & 0xf) << 8)
-#define BCM2838_DMA40_INC		BIT(12)
-#define BCM2838_DMA40_SIZE_128	(2 << 13)
+/* the max dma length for different channels */
+#define MAX_DMA40_LEN SZ_1G
 
-#define BCM2838_DMA40_MEMCPY_QOS \
-	(BCM2838_DMA40_CS_QOS(0x0) | \
-	 BCM2838_DMA40_CS_PANIC_QOS(0x0) | \
-	 BCM2838_DMA40_CS_WRITE_WAIT)
+#define BCM2838_DMA40_BURST_LEN(x)	((min(x,16) - 1) << 8)
+#define BCM2838_DMA40_INC		BIT(12)
+#define BCM2838_DMA40_SIZE_32		(0 << 13)
+#define BCM2838_DMA40_SIZE_64		(1 << 13)
+#define BCM2838_DMA40_SIZE_128		(2 << 13)
+#define BCM2838_DMA40_SIZE_256		(3 << 13)
+#define BCM2838_DMA40_IGNORE		BIT(15)
+#define BCM2838_DMA40_STRIDE(x)		((x) << 16) /* For 2D mode */
+
+#define BCM2838_DMA40_MEMCPY_FLAGS \
+	(BCM2838_DMA40_QOS(0) | \
+	 BCM2838_DMA40_PANIC_QOS(0) | \
+	 BCM2838_DMA40_WAIT_FOR_WRITES | \
+	 BCM2838_DMA40_DISDEBUG)
 
 #define BCM2838_DMA40_MEMCPY_XFER_INFO \
 	(BCM2838_DMA40_SIZE_128 | \
 	 BCM2838_DMA40_INC | \
 	 BCM2838_DMA40_BURST_LEN(16))
 
+struct bcm2835_dmadev *memcpy_parent;
 static void __iomem *memcpy_chan;
 static struct bcm2838_dma40_scb *memcpy_scb;
 static dma_addr_t memcpy_scb_dma;
 DEFINE_SPINLOCK(memcpy_lock);
 
+static const struct bcm2835_dma_cfg_data bcm2835_dma_cfg = {
+	.chan_40bit_mask = 0,
+};
+
+static const struct bcm2835_dma_cfg_data bcm2838_dma_cfg = {
+	.chan_40bit_mask = BIT(11) | BIT(12) | BIT(13) | BIT(14),
+};
+
 static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
 {
 	/* lite and normal channels have different max frame length */
@@ -264,6 +337,32 @@ static inline struct bcm2835_desc *to_bc
 	return container_of(t, struct bcm2835_desc, vd.tx);
 }
 
+static inline uint32_t to_bcm2838_ti(uint32_t info)
+{
+	return ((info & BCM2835_DMA_INT_EN) ? BCM2838_DMA40_INTEN : 0) |
+		((info & BCM2835_DMA_WAIT_RESP) ? BCM2838_DMA40_WAIT_RESP : 0) |
+		((info & BCM2835_DMA_S_DREQ) ?
+		 (BCM2838_DMA40_S_DREQ | BCM2838_DMA40_WAIT_RD_RESP) : 0) |
+		((info & BCM2835_DMA_D_DREQ) ? BCM2838_DMA40_D_DREQ : 0) |
+		BCM2838_DMA40_PER_MAP((info >> 16) & 0x1f);
+}
+
+static inline uint32_t to_bcm2838_srci(uint32_t info)
+{
+	return ((info & BCM2835_DMA_S_INC) ? BCM2838_DMA40_INC : 0);
+}
+
+static inline uint32_t to_bcm2838_dsti(uint32_t info)
+{
+	return ((info & BCM2835_DMA_D_INC) ? BCM2838_DMA40_INC : 0);
+}
+
+static inline uint32_t to_bcm2838_cbaddr(dma_addr_t addr)
+{
+	BUG_ON(addr & 0x1f);
+	return (addr >> 5);
+}
+
 static void bcm2835_dma_free_cb_chain(struct bcm2835_desc *desc)
 {
 	size_t i;
@@ -282,45 +381,53 @@ static void bcm2835_dma_desc_free(struct
 }
 
 static void bcm2835_dma_create_cb_set_length(
-	struct bcm2835_chan *chan,
+	struct bcm2835_chan *c,
 	struct bcm2835_dma_cb *control_block,
 	size_t len,
 	size_t period_len,
 	size_t *total_len,
 	u32 finalextrainfo)
 {
-	size_t max_len = bcm2835_dma_max_frame_length(chan);
+	size_t max_len = bcm2835_dma_max_frame_length(c);
+	uint32_t cb_len;
 
 	/* set the length taking lite-channel limitations into account */
-	control_block->length = min_t(u32, len, max_len);
+	cb_len = min_t(u32, len, max_len);
 
-	/* finished if we have no period_length */
-	if (!period_len)
-		return;
+	if (period_len) {
+		/*
+		 * 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
+		 */
 
-	/*
-	 * 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 + cb_len < period_len) {
+			/* update number of bytes in this period so far */
+			*total_len += cb_len;
+		} else {
+			/* calculate the length that remains to reach period_len */
+			cb_len = period_len - *total_len;
 
-	/* have we filled in period_length yet? */
-	if (*total_len + control_block->length < period_len) {
-		/* update number of bytes in this period so far */
-		*total_len += control_block->length;
-		return;
+			/* reset total_length for next period */
+			*total_len = 0;
+		}
 	}
 
-	/* 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;
+	if (c->is_40bit_channel) {
+		struct bcm2838_dma40_scb *scb =
+			(struct bcm2838_dma40_scb *)control_block;
+
+		scb->len = cb_len;
+		/* add extrainfo bits to ti */
+		scb->ti |= to_bcm2838_ti(finalextrainfo);
+	} else {
+		control_block->length = cb_len;
+		/* add extrainfo bits to info */
+		control_block->info |= finalextrainfo;
+	}
 }
 
 static inline size_t bcm2835_dma_count_frames_for_sg(
@@ -343,7 +450,7 @@ static inline size_t bcm2835_dma_count_f
 /**
  * bcm2835_dma_create_cb_chain - create a control block and fills data in
  *
- * @chan:           the @dma_chan for which we run this
+ * @c:              the @bcm2835_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
@@ -361,12 +468,11 @@ static inline size_t bcm2835_dma_count_f
  * @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,
+	struct bcm2835_chan *c, 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;
@@ -399,11 +505,23 @@ static struct bcm2835_desc *bcm2835_dma_
 
 		/* 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;
+		if (c->is_40bit_channel) {
+			struct bcm2838_dma40_scb *scb =
+				(struct bcm2838_dma40_scb *)control_block;
+			scb->ti = to_bcm2838_ti(info);
+			scb->src = lower_32_bits(src);
+			scb->srci= upper_32_bits(src) | to_bcm2838_srci(info);
+			scb->dst = lower_32_bits(dst);
+			scb->dsti = upper_32_bits(dst) | to_bcm2838_dsti(info);
+			scb->next_cb = 0;
+		} else {
+			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 */
@@ -417,7 +535,10 @@ static struct bcm2835_desc *bcm2835_dma_
 		}
 
 		/* link this the last controlblock */
-		if (frame)
+		if (frame && c->is_40bit_channel)
+			d->cb_list[frame - 1].cb->next =
+				to_bcm2838_cbaddr(cb_entry->paddr);
+		if (frame && !c->is_40bit_channel)
 			d->cb_list[frame - 1].cb->next = cb_entry->paddr;
 
 		/* update src and dst and length */
@@ -431,7 +552,14 @@ static struct bcm2835_desc *bcm2835_dma_
 	}
 
 	/* the last frame requires extra flags */
-	d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
+	if (c->is_40bit_channel) {
+		struct bcm2838_dma40_scb *scb =
+			(struct bcm2838_dma40_scb *)d->cb_list[d->frames-1].cb;
+
+		scb->ti |= to_bcm2838_ti(finalextrainfo);
+	} else {
+		d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
+	}
 
 	/* detect a size missmatch */
 	if (buf_len && (d->size != buf_len))
@@ -445,28 +573,51 @@ error_cb:
 }
 
 static void bcm2835_dma_fill_cb_chain_with_sg(
-	struct dma_chan *chan,
+	struct bcm2835_chan *c,
 	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 len, max_len;
 	unsigned int i;
 	dma_addr_t addr;
 	struct scatterlist *sgent;
 
+	pr_err("dma_fill_chain_with_sg(ch %d, dir %d):\n", c->ch, direction);
+
 	max_len = bcm2835_dma_max_frame_length(c);
 	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);
+		if (c->is_40bit_channel) {
+			struct bcm2838_dma40_scb *scb =
+				(struct bcm2838_dma40_scb *)cb->cb;
+			for (addr = sg_dma_address(sgent),
+				     len = sg_dma_len(sgent);
+			     len > 0;
+			     addr += scb->len, len -= scb->len, scb++) {
+				if (direction == DMA_DEV_TO_MEM) {
+					scb->dst = lower_32_bits(addr);
+					scb->dsti = upper_32_bits(addr) | BCM2838_DMA40_INC;
+				} else {
+					scb->src = lower_32_bits(addr);
+					scb->srci = upper_32_bits(addr) | BCM2838_DMA40_INC;
+				}
+				scb->len = min(len, max_len);
+				pr_err("  %llx, %x\n", (u64)addr, scb->len);
+			}
+		} else {
+			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);
+				pr_err("  %llx, %x\n", (u64)addr, cb->cb->length);
+			}
 		}
 	}
 }
@@ -475,6 +626,10 @@ static int bcm2835_dma_abort(struct bcm2
 {
 	void __iomem *chan_base = c->chan_base;
 	long int timeout = 10000;
+	u32 wait_mask = BCM2835_DMA_WAITING_FOR_WRITES;
+
+	if (c->is_40bit_channel)
+		wait_mask = BCM2838_DMA40_WAITING_FOR_WRITES;
 
 	/*
 	 * A zero control block address means the channel is idle.
@@ -487,8 +642,7 @@ static int bcm2835_dma_abort(struct bcm2
 	writel(0, chan_base + BCM2835_DMA_CS);
 
 	/* Wait for any current AXI transfer to complete */
-	while ((readl(chan_base + BCM2835_DMA_CS) &
-		BCM2835_DMA_WAITING_FOR_WRITES) && --timeout)
+	while ((readl(chan_base + BCM2835_DMA_CS) & wait_mask) && --timeout)
 		cpu_relax();
 
 	/* Peripheral might be stuck and fail to signal AXI write responses */
@@ -505,6 +659,7 @@ static void bcm2835_dma_start_desc(struc
 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
 	struct bcm2835_desc *d;
 
+	pr_err("dma_start_desc(%px)\n", vd);
 	if (!vd) {
 		c->desc = NULL;
 		return;
@@ -514,9 +669,16 @@ static void bcm2835_dma_start_desc(struc
 
 	c->desc = d = to_bcm2835_dma_desc(&vd->tx);
 
-	writel(d->cb_list[0].paddr, c->chan_base + BCM2835_DMA_ADDR);
-	writel(BCM2835_DMA_ACTIVE | BCM2835_DMA_CS_FLAGS(c->dreq),
-	       c->chan_base + BCM2835_DMA_CS);
+	if (c->is_40bit_channel) {
+		writel(to_bcm2838_cbaddr(d->cb_list[0].paddr),
+		       c->chan_base + BCM2838_DMA40_CB);
+		writel(BCM2838_DMA40_ACTIVE | BCM2838_DMA40_CS_FLAGS(c->dreq),
+		       c->chan_base + BCM2838_DMA40_CS);
+	} else {
+		writel(d->cb_list[0].paddr, c->chan_base + BCM2835_DMA_ADDR);
+		writel(BCM2835_DMA_ACTIVE | BCM2835_DMA_CS_FLAGS(c->dreq),
+		       c->chan_base + BCM2835_DMA_CS);
+	}
 }
 
 static irqreturn_t bcm2835_dma_callback(int irq, void *data)
@@ -544,7 +706,8 @@ static irqreturn_t bcm2835_dma_callback(
 	 * will remain idle despite the ACTIVE flag being set.
 	 */
 	writel(BCM2835_DMA_INT | BCM2835_DMA_ACTIVE |
-	       BCM2835_DMA_CS_FLAGS(c->dreq),
+	       (c->is_40bit_channel ? BCM2838_DMA40_CS_FLAGS(c->dreq) :
+		BCM2835_DMA_CS_FLAGS(c->dreq)),
 	       c->chan_base + BCM2835_DMA_CS);
 
 	d = c->desc;
@@ -643,9 +806,17 @@ static enum dma_status bcm2835_dma_tx_st
 		struct bcm2835_desc *d = c->desc;
 		dma_addr_t pos;
 
-		if (d->dir == DMA_MEM_TO_DEV)
+		if (d->dir == DMA_MEM_TO_DEV && c->is_40bit_channel)
+			pos = readl(c->chan_base + BCM2838_DMA40_SRC) +
+				((readl(c->chan_base + BCM2838_DMA40_SRCI) &
+				  0xff) << 8);
+		else if (d->dir == DMA_MEM_TO_DEV && !c->is_40bit_channel)
 			pos = readl(c->chan_base + BCM2835_DMA_SOURCE_AD);
-		else if (d->dir == DMA_DEV_TO_MEM)
+		else if (d->dir == DMA_DEV_TO_MEM && c->is_40bit_channel)
+			pos = readl(c->chan_base + BCM2838_DMA40_DEST) +
+				((readl(c->chan_base + BCM2838_DMA40_DESTI) &
+				  0xff) << 8);
+		else if (d->dir == DMA_DEV_TO_MEM && !c->is_40bit_channel)
 			pos = readl(c->chan_base + BCM2835_DMA_DEST_AD);
 		else
 			pos = 0;
@@ -691,7 +862,7 @@ static struct dma_async_tx_descriptor *b
 	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,
+	d = bcm2835_dma_create_cb_chain(c, DMA_MEM_TO_MEM, false,
 					info, extra, frames,
 					src, dst, len, 0, GFP_KERNEL);
 	if (!d)
@@ -726,11 +897,21 @@ static struct dma_async_tx_descriptor *b
 		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
 			return NULL;
 		src = c->cfg.src_addr;
+		/*
+		 * One would think it ought to be possible to get the physical
+		 * to dma address mapping information from the dma-ranges DT
+		 * property, but I've not found a way yet that doesn't involve
+		 * open-coding the whole thing.
+		 */
+		if (c->is_40bit_channel)
+		    src |= 0x400000000ull;
 		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;
+		if (c->is_40bit_channel)
+		    dst |= 0x400000000ull;
 		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
 	}
 
@@ -738,7 +919,7 @@ static struct dma_async_tx_descriptor *b
 	frames = bcm2835_dma_count_frames_for_sg(c, sgl, sg_len);
 
 	/* allocate the CB chain */
-	d = bcm2835_dma_create_cb_chain(chan, direction, false,
+	d = bcm2835_dma_create_cb_chain(c, direction, false,
 					info, extra,
 					frames, src, dst, 0, 0,
 					GFP_KERNEL);
@@ -746,7 +927,7 @@ static struct dma_async_tx_descriptor *b
 		return NULL;
 
 	/* fill in frames with scatterlist pointers */
-	bcm2835_dma_fill_cb_chain_with_sg(chan, direction, d->cb_list,
+	bcm2835_dma_fill_cb_chain_with_sg(c, direction, d->cb_list,
 					  sgl, sg_len);
 
 	return vchan_tx_prep(&c->vc, &d->vd, flags);
@@ -815,7 +996,7 @@ static struct dma_async_tx_descriptor *b
 	 * note that we need to use GFP_NOWAIT, as the ALSA i2s dmaengine
 	 * implementation calls prep_dma_cyclic with interrupts disabled.
 	 */
-	d = bcm2835_dma_create_cb_chain(chan, direction, true,
+	d = bcm2835_dma_create_cb_chain(c, direction, true,
 					info, extra,
 					frames, src, dst, buf_len,
 					period_len, GFP_NOWAIT);
@@ -823,7 +1004,8 @@ static struct dma_async_tx_descriptor *b
 		return NULL;
 
 	/* wrap around into a loop */
-	d->cb_list[d->frames - 1].cb->next = d->cb_list[0].paddr;
+	d->cb_list[d->frames - 1].cb->next = c->is_40bit_channel ?
+		to_bcm2838_cbaddr(d->cb_list[0].paddr) : d->cb_list[0].paddr;
 
 	return vchan_tx_prep(&c->vc, &d->vd, flags);
 }
@@ -899,9 +1081,11 @@ static int bcm2835_dma_chan_init(struct
 	c->irq_number = irq;
 	c->irq_flags = irq_flags;
 
-	/* check in DEBUG register if this is a LITE channel */
-	if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
-		BCM2835_DMA_DEBUG_LITE)
+	/* check for 40bit and lite channels */
+	if (d->cfg_data->chan_40bit_mask & BIT(chan_id))
+		c->is_40bit_channel = true;
+	else if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
+		 BCM2835_DMA_DEBUG_LITE)
 		c->is_lite_channel = true;
 
 	return 0;
@@ -918,18 +1102,16 @@ static void bcm2835_dma_free(struct bcm2
 	}
 }
 
-int bcm2838_dma40_memcpy_init(struct device *dev)
+int bcm2838_dma40_memcpy_init(void)
 {
-	if (memcpy_scb)
-		return 0;
+	if (!memcpy_parent)
+		return -EPROBE_DEFER;
 
-	memcpy_scb = dma_alloc_coherent(dev, sizeof(*memcpy_scb),
-					&memcpy_scb_dma, GFP_KERNEL);
+	if (!memcpy_chan)
+		return -EINVAL;
 
-	if (!memcpy_scb) {
-		pr_err("bcm2838_dma40_memcpy_init failed!\n");
+	if (!memcpy_scb)
 		return -ENOMEM;
-	}
 
 	return 0;
 }
@@ -956,20 +1138,22 @@ void bcm2838_dma40_memcpy(dma_addr_t dst
 	scb->next_cb = 0;
 
 	writel((u32)(memcpy_scb_dma >> 5), memcpy_chan + BCM2838_DMA40_CB);
-	writel(BCM2838_DMA40_MEMCPY_QOS + BCM2838_DMA40_CS_ACTIVE,
+	writel(BCM2838_DMA40_MEMCPY_FLAGS + BCM2838_DMA40_ACTIVE,
 	       memcpy_chan + BCM2838_DMA40_CS);
+
 	/* Poll for completion */
-	while (!(readl(memcpy_chan + BCM2838_DMA40_CS) & BCM2838_DMA40_CS_END))
+	while (!(readl(memcpy_chan + BCM2838_DMA40_CS) & BCM2838_DMA40_END))
 		cpu_relax();
 
-	writel(BCM2838_DMA40_CS_END, memcpy_chan + BCM2838_DMA40_CS);
+	writel(BCM2838_DMA40_END, memcpy_chan + BCM2838_DMA40_CS);
 
 	spin_unlock_irqrestore(&memcpy_lock, flags);
 }
 EXPORT_SYMBOL(bcm2838_dma40_memcpy);
 
 static const struct of_device_id bcm2835_dma_of_match[] = {
-	{ .compatible = "brcm,bcm2835-dma", },
+	{ .compatible = "brcm,bcm2835-dma", .data = &bcm2835_dma_cfg },
+	{ .compatible = "brcm,bcm2838-dma", .data = &bcm2838_dma_cfg },
 	{},
 };
 MODULE_DEVICE_TABLE(of, bcm2835_dma_of_match);
@@ -1001,6 +1185,8 @@ static int bcm2835_dma_probe(struct plat
 	int irq_flags;
 	uint32_t chans_available;
 	char chan_name[BCM2835_DMA_CHAN_NAME_SIZE];
+	const struct of_device_id *of_id;
+	int chan_count, chan_start, chan_end;
 
 	if (!pdev->dev.dma_mask)
 		pdev->dev.dma_mask = &pdev->dev.coherent_dma_mask;
@@ -1022,9 +1208,13 @@ static int bcm2835_dma_probe(struct plat
 	base = devm_ioremap_resource(&pdev->dev, res);
 	if (IS_ERR(base))
 		return PTR_ERR(base);
-	rc = bcm_dmaman_probe(pdev, base, BCM2835_DMA_BULK_MASK);
-	if (rc)
-		dev_err(&pdev->dev, "Failed to initialize the legacy API\n");
+
+	/* The set of channels can be split across multiple instances. */
+	chan_start = ((u32)base / BCM2835_DMA_CHAN_SIZE) & 0xf;
+	base -= BCM2835_DMA_CHAN(chan_start);
+	chan_count = resource_size(res) / BCM2835_DMA_CHAN_SIZE;
+	chan_end = min(chan_start + chan_count,
+			 BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED + 1);
 
 	od->base = base;
 
@@ -1054,6 +1244,14 @@ static int bcm2835_dma_probe(struct plat
 
 	platform_set_drvdata(pdev, od);
 
+	of_id = of_match_node(bcm2835_dma_of_match, pdev->dev.of_node);
+	if (!of_id) {
+		dev_err(&pdev->dev, "Failed to match compatible string\n");
+		return -EINVAL;
+	}
+
+	od->cfg_data = of_id->data;
+
 	/* Request DMA channel mask from device tree */
 	if (of_property_read_u32(pdev->dev.of_node,
 			"brcm,dma-channel-mask",
@@ -1063,18 +1261,34 @@ static int bcm2835_dma_probe(struct plat
 		goto err_no_dma;
 	}
 
-	/* Channel 0 is used by the legacy API */
-	chans_available &= ~BCM2835_DMA_BULK_MASK;
+	/* One channel is reserved for the legacy API */
+	if (chans_available & BCM2835_DMA_BULK_MASK) {
+		rc = bcm_dmaman_probe(pdev, base,
+				      chans_available & BCM2835_DMA_BULK_MASK);
+		if (rc)
+			dev_err(&pdev->dev,
+				"Failed to initialize the legacy API\n");
+
+		chans_available &= ~BCM2835_DMA_BULK_MASK;
+	}
 
-	/* We can't use channels 11-13 yet */
-	chans_available &= ~(BIT(11) | BIT(12) | BIT(13));
+	/* And possibly one for the 40-bit DMA memcpy API */
+	if (chans_available & od->cfg_data->chan_40bit_mask &
+	    BIT(BCM2838_DMA_MEMCPY_CHAN)) {
+		memcpy_parent = od;
+		memcpy_chan = BCM2835_DMA_CHANIO(base, BCM2838_DMA_MEMCPY_CHAN);
+		memcpy_scb = dma_alloc_coherent(memcpy_parent->ddev.dev,
+						sizeof(*memcpy_scb),
+						&memcpy_scb_dma, GFP_KERNEL);
+		if (!memcpy_scb)
+			dev_warn(&pdev->dev,
+				 "Failed to allocated memcpy scb\n");
 
-	/* Grab channel 14 for the 40-bit DMA memcpy */
-	chans_available &= ~BIT(14);
-	memcpy_chan = BCM2835_DMA_CHANIO(base, 14);
+		chans_available &= ~BIT(BCM2838_DMA_MEMCPY_CHAN);
+	}
 
 	/* get irqs for each channel that we support */
-	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
+	for (i = chan_start; i < chan_end; i++) {
 		/* skip masked out channels */
 		if (!(chans_available & (1 << i))) {
 			irq[i] = -1;
@@ -1097,13 +1311,17 @@ static int bcm2835_dma_probe(struct plat
 		irq[i] = platform_get_irq(pdev, i < 11 ? i : 11);
 	}
 
+	chan_count = 0;
+
 	/* get irqs for each channel */
-	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
+	for (i = chan_start; i < chan_end; i++) {
 		/* skip channels without irq */
 		if (irq[i] < 0)
 			continue;
 
 		/* check if there are other channels that also use this irq */
+		/* FIXME: This will fail if interrupts are shared across
+		   instances */
 		irq_flags = 0;
 		for (j = 0; j <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; j++)
 			if ((i != j) && (irq[j] == irq[i])) {
@@ -1115,9 +1333,10 @@ static int bcm2835_dma_probe(struct plat
 		rc = bcm2835_dma_chan_init(od, i, irq[i], irq_flags);
 		if (rc)
 			goto err_no_dma;
+		chan_count++;
 	}
 
-	dev_dbg(&pdev->dev, "Initialized %i DMA channels\n", i);
+	dev_dbg(&pdev->dev, "Initialized %i DMA channels\n", chan_count);
 
 	/* Device-tree DMA controller registration */
 	rc = of_dma_controller_register(pdev->dev.of_node,
@@ -1149,6 +1368,13 @@ static int bcm2835_dma_remove(struct pla
 
 	bcm_dmaman_remove(pdev);
 	dma_async_device_unregister(&od->ddev);
+	if (memcpy_parent == od) {
+		dma_free_coherent(&pdev->dev, sizeof(*memcpy_scb), memcpy_scb,
+				  memcpy_scb_dma);
+		memcpy_parent = NULL;
+		memcpy_scb = NULL;
+		memcpy_chan = NULL;
+	}
 	bcm2835_dma_free(od);
 
 	return 0;
--- a/drivers/pci/controller/pcie-brcmstb-bounce.c
+++ b/drivers/pci/controller/pcie-brcmstb-bounce.c
@@ -91,7 +91,7 @@ struct dmabounce_device_info {
 
 static struct dmabounce_device_info *g_dmabounce_device_info;
 
-extern int bcm2838_dma40_memcpy_init(struct device *dev);
+extern int bcm2838_dma40_memcpy_init(void);
 extern void bcm2838_dma40_memcpy(dma_addr_t dst, dma_addr_t src, size_t size);
 
 #ifdef STATS
@@ -471,9 +471,9 @@ static const struct dma_map_ops dmabounc
 	.mapping_error		= dmabounce_mapping_error,
 };
 
-int brcm_pcie_bounce_register_dev(struct device *dev,
-				  unsigned long buffer_size,
-				  dma_addr_t threshold)
+int brcm_pcie_bounce_init(struct device *dev,
+			  unsigned long buffer_size,
+			  dma_addr_t threshold)
 {
 	struct dmabounce_device_info *device_info;
 	int ret;
@@ -482,9 +482,9 @@ int brcm_pcie_bounce_register_dev(struct
 	if (g_dmabounce_device_info)
 		return -EBUSY;
 
-	ret = bcm2838_dma40_memcpy_init(dev);
+	ret = bcm2838_dma40_memcpy_init();
 	if (ret)
-	    return ret;
+		return ret;
 
 	device_info = kmalloc(sizeof(struct dmabounce_device_info), GFP_ATOMIC);
 	if (!device_info) {
@@ -515,9 +515,8 @@ int brcm_pcie_bounce_register_dev(struct
 		 device_create_file(dev, &dev_attr_dmabounce_stats));
 
 	g_dmabounce_device_info = device_info;
-	set_dma_ops(dev, &dmabounce_ops);
 
-	dev_info(dev, "dmabounce: registered device - %ld kB, threshold %pad\n",
+	dev_info(dev, "dmabounce: initialised - %ld kB, threshold %pad\n",
 		 buffer_size / 1024, &threshold);
 
 	return 0;
@@ -526,14 +525,13 @@ int brcm_pcie_bounce_register_dev(struct
 	kfree(device_info);
 	return ret;
 }
-EXPORT_SYMBOL(brcm_pcie_bounce_register_dev);
+EXPORT_SYMBOL(brcm_pcie_bounce_init);
 
-void brcm_pcie_bounce_unregister_dev(struct device *dev)
+void brcm_pcie_bounce_uninit(struct device *dev)
 {
 	struct dmabounce_device_info *device_info = g_dmabounce_device_info;
 
 	g_dmabounce_device_info = NULL;
-	set_dma_ops(dev, NULL);
 
 	if (!device_info) {
 		dev_warn(dev,
@@ -554,10 +552,16 @@ void brcm_pcie_bounce_unregister_dev(str
 			 device_remove_file(dev, &dev_attr_dmabounce_stats));
 
 	kfree(device_info);
+}
+EXPORT_SYMBOL(brcm_pcie_bounce_uninit);
+
+int brcm_pcie_bounce_register_dev(struct device *dev)
+{
+	set_dma_ops(dev, &dmabounce_ops);
 
-	dev_info(dev, "dmabounce: device unregistered\n");
+	return 0;
 }
-EXPORT_SYMBOL(brcm_pcie_bounce_unregister_dev);
+EXPORT_SYMBOL(brcm_pcie_bounce_register_dev);
 
 MODULE_AUTHOR("Phil Elwell <phil@raspberrypi.org>");
 MODULE_DESCRIPTION("Dedicate DMA bounce support for pcie-brcmstb");
--- a/drivers/pci/controller/pcie-brcmstb-bounce.h
+++ b/drivers/pci/controller/pcie-brcmstb-bounce.h
@@ -8,21 +8,26 @@
 
 #ifdef CONFIG_ARM
 
-int brcm_pcie_bounce_register_dev(struct device *dev, unsigned long buffer_size,
-				  dma_addr_t threshold);
-
-int brcm_pcie_bounce_unregister_dev(struct device *dev);
+int brcm_pcie_bounce_init(struct device *dev, unsigned long buffer_size,
+			  dma_addr_t threshold);
+int brcm_pcie_bounce_uninit(struct device *dev);
+int brcm_pcie_bounce_register_dev(struct device *dev);
 
 #else
 
-static inline int brcm_pcie_bounce_register_dev(struct device *dev,
-						unsigned long buffer_size,
-						dma_addr_t threshold)
+static inline int brcm_pcie_bounce_init(struct device *dev,
+					unsigned long buffer_size,
+					dma_addr_t threshold)
+{
+	return 0;
+}
+
+static inline int brcm_pcie_bounce_uninit(struct device *dev)
 {
 	return 0;
 }
 
-static inline int brcm_pcie_bounce_unregister_dev(struct device *dev)
+static inline int brcm_pcie_bounce_register_dev(struct device *dev)
 {
 	return 0;
 }
--- a/drivers/pci/controller/pcie-brcmstb.c
+++ b/drivers/pci/controller/pcie-brcmstb.c
@@ -650,6 +650,7 @@ static void brcm_set_dma_ops(struct devi
 
 static inline void brcm_pcie_perst_set(struct brcm_pcie *pcie,
 				       unsigned int val);
+
 static int brcmstb_platform_notifier(struct notifier_block *nb,
 				     unsigned long event, void *__dev)
 {
@@ -663,12 +664,11 @@ static int brcmstb_platform_notifier(str
 		    strcmp(dev->kobj.name, rc_name)) {
 			int ret;
 
-			ret = brcm_pcie_bounce_register_dev(dev, bounce_buffer,
-							    (dma_addr_t)bounce_threshold);
+			ret = brcm_pcie_bounce_register_dev(dev);
 			if (ret) {
 				dev_err(dev,
 					"brcm_pcie_bounce_register_dev() failed: %d\n",
-				ret);
+					ret);
 				return ret;
 			}
 		}
@@ -681,8 +681,6 @@ static int brcmstb_platform_notifier(str
 			brcm_pcie_perst_set(g_pcie, 1);
 			msleep(100);
 			brcm_pcie_perst_set(g_pcie, 0);
-		} else if (max_pfn > (bounce_threshold/PAGE_SIZE)) {
-			brcm_pcie_bounce_unregister_dev(dev);
 		}
 		return NOTIFY_OK;
 
@@ -1718,6 +1716,7 @@ static int brcm_pcie_probe(struct platfo
 	void __iomem *base;
 	struct pci_host_bridge *bridge;
 	struct pci_bus *child;
+	extern unsigned long max_pfn;
 
 	bridge = devm_pci_alloc_host_bridge(&pdev->dev, sizeof(*pcie));
 	if (!bridge)
@@ -1753,6 +1752,20 @@ static int brcm_pcie_probe(struct platfo
 	if (IS_ERR(base))
 		return PTR_ERR(base);
 
+	/* To Do: Add hardware check if this ever gets fixed */
+	if (max_pfn > (bounce_threshold/PAGE_SIZE)) {
+		int ret;
+		ret = brcm_pcie_bounce_init(&pdev->dev, bounce_buffer,
+					    (dma_addr_t)bounce_threshold);
+		if (ret) {
+			if (ret != -EPROBE_DEFER)
+				dev_err(&pdev->dev,
+					"could not init bounce buffers: %d\n",
+					ret);
+			return ret;
+		}
+	}
+
 	pcie->clk = of_clk_get_by_name(dn, "sw_pcie");
 	if (IS_ERR(pcie->clk)) {
 		dev_warn(&pdev->dev, "could not get clock\n");