dmaengine: axi-dmac: Add support for scatter-gather transfers

Implement support for scatter-gather transfers. Build a chain of
hardware descriptors, each one corresponding to a segment of the
transfer, and linked to the next one. The hardware will transfer the
chain and only fire interrupts when the whole chain has been
transferred.

Support for scatter-gather is automatically enabled when the driver
detects that the hardware supports it, by writing then reading the
AXI_DMAC_REG_SG_ADDRESS register. If not available, the driver will fall
back to standard DMA transfers.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Link: https://lore.kernel.org/r/20231215131313.23840-4-paul@crapouillou.net
Signed-off-by: Vinod Koul <vkoul@kernel.org>

Author: pcercuei

drivers/dma/dma-axi-dmac.c

@@ -84,9 +84,13 @@
 #define AXI_DMAC_REG_DBG2		0x444
 #define AXI_DMAC_REG_PARTIAL_XFER_LEN	0x44c
 #define AXI_DMAC_REG_PARTIAL_XFER_ID	0x450
+#define AXI_DMAC_REG_CURRENT_SG_ID	0x454
+#define AXI_DMAC_REG_SG_ADDRESS		0x47c
+#define AXI_DMAC_REG_SG_ADDRESS_HIGH	0x4bc
 
 #define AXI_DMAC_CTRL_ENABLE		BIT(0)
 #define AXI_DMAC_CTRL_PAUSE		BIT(1)
+#define AXI_DMAC_CTRL_ENABLE_SG		BIT(2)
 
 #define AXI_DMAC_IRQ_SOT		BIT(0)
 #define AXI_DMAC_IRQ_EOT		BIT(1)
@@ -100,12 +104,16 @@
 /* The maximum ID allocated by the hardware is 31 */
 #define AXI_DMAC_SG_UNUSED 32U
 
+/* Flags for axi_dmac_hw_desc.flags */
+#define AXI_DMAC_HW_FLAG_LAST		BIT(0)
+#define AXI_DMAC_HW_FLAG_IRQ		BIT(1)
+
 struct axi_dmac_hw_desc {
 	u32 flags;
 	u32 id;
 	u64 dest_addr;
 	u64 src_addr;
-	u64 __unused;
+	u64 next_sg_addr;
 	u32 y_len;
 	u32 x_len;
 	u32 src_stride;
@@ -153,6 +161,7 @@ struct axi_dmac_chan {
 	bool hw_partial_xfer;
 	bool hw_cyclic;
 	bool hw_2d;
+	bool hw_sg;
 };
 
 struct axi_dmac {
@@ -227,9 +236,11 @@ static void axi_dmac_start_transfer(struct axi_dmac_chan *chan)
 	unsigned int flags = 0;
 	unsigned int val;
 
-	val = axi_dmac_read(dmac, AXI_DMAC_REG_START_TRANSFER);
-	if (val) /* Queue is full, wait for the next SOT IRQ */
-		return;
+	if (!chan->hw_sg) {
+		val = axi_dmac_read(dmac, AXI_DMAC_REG_START_TRANSFER);
+		if (val) /* Queue is full, wait for the next SOT IRQ */
+			return;
+	}
 
 	desc = chan->next_desc;
 
@@ -248,9 +259,10 @@ static void axi_dmac_start_transfer(struct axi_dmac_chan *chan)
 		return;
 	}
 
-	desc->num_submitted++;
-	if (desc->num_submitted == desc->num_sgs ||
-	    desc->have_partial_xfer) {
+	if (chan->hw_sg) {
+		chan->next_desc = NULL;
+	} else if (++desc->num_submitted == desc->num_sgs ||
+		   desc->have_partial_xfer) {
 		if (desc->cyclic)
 			desc->num_submitted = 0; /* Start again */
 		else
@@ -262,14 +274,16 @@ static void axi_dmac_start_transfer(struct axi_dmac_chan *chan)
 
 	sg->hw->id = axi_dmac_read(dmac, AXI_DMAC_REG_TRANSFER_ID);
 
-	if (axi_dmac_dest_is_mem(chan)) {
-		axi_dmac_write(dmac, AXI_DMAC_REG_DEST_ADDRESS, sg->hw->dest_addr);
-		axi_dmac_write(dmac, AXI_DMAC_REG_DEST_STRIDE, sg->hw->dst_stride);
-	}
+	if (!chan->hw_sg) {
+		if (axi_dmac_dest_is_mem(chan)) {
+			axi_dmac_write(dmac, AXI_DMAC_REG_DEST_ADDRESS, sg->hw->dest_addr);
+			axi_dmac_write(dmac, AXI_DMAC_REG_DEST_STRIDE, sg->hw->dst_stride);
+		}
 
-	if (axi_dmac_src_is_mem(chan)) {
-		axi_dmac_write(dmac, AXI_DMAC_REG_SRC_ADDRESS, sg->hw->src_addr);
-		axi_dmac_write(dmac, AXI_DMAC_REG_SRC_STRIDE, sg->hw->src_stride);
+		if (axi_dmac_src_is_mem(chan)) {
+			axi_dmac_write(dmac, AXI_DMAC_REG_SRC_ADDRESS, sg->hw->src_addr);
+			axi_dmac_write(dmac, AXI_DMAC_REG_SRC_STRIDE, sg->hw->src_stride);
+		}
 	}
 
 	/*
@@ -284,8 +298,14 @@ static void axi_dmac_start_transfer(struct axi_dmac_chan *chan)
 	if (chan->hw_partial_xfer)
 		flags |= AXI_DMAC_FLAG_PARTIAL_REPORT;
 
-	axi_dmac_write(dmac, AXI_DMAC_REG_X_LENGTH, sg->hw->x_len);
-	axi_dmac_write(dmac, AXI_DMAC_REG_Y_LENGTH, sg->hw->y_len);
+	if (chan->hw_sg) {
+		axi_dmac_write(dmac, AXI_DMAC_REG_SG_ADDRESS, (u32)sg->hw_phys);
+		axi_dmac_write(dmac, AXI_DMAC_REG_SG_ADDRESS_HIGH,
+			       (u64)sg->hw_phys >> 32);
+	} else {
+		axi_dmac_write(dmac, AXI_DMAC_REG_X_LENGTH, sg->hw->x_len);
+		axi_dmac_write(dmac, AXI_DMAC_REG_Y_LENGTH, sg->hw->y_len);
+	}
 	axi_dmac_write(dmac, AXI_DMAC_REG_FLAGS, flags);
 	axi_dmac_write(dmac, AXI_DMAC_REG_START_TRANSFER, 1);
 }
@@ -362,6 +382,9 @@ static void axi_dmac_compute_residue(struct axi_dmac_chan *chan,
 	rslt->result = DMA_TRANS_NOERROR;
 	rslt->residue = 0;
 
+	if (chan->hw_sg)
+		return;
+
 	/*
 	 * We get here if the last completed segment is partial, which
 	 * means we can compute the residue from that segment onwards
@@ -388,36 +411,46 @@ static bool axi_dmac_transfer_done(struct axi_dmac_chan *chan,
 	    (completed_transfers & AXI_DMAC_FLAG_PARTIAL_XFER_DONE))
 		axi_dmac_dequeue_partial_xfers(chan);
 
-	do {
-		sg = &active->sg[active->num_completed];
-		if (sg->hw->id == AXI_DMAC_SG_UNUSED) /* Not yet submitted */
-			break;
-		if (!(BIT(sg->hw->id) & completed_transfers))
-			break;
-		active->num_completed++;
-		sg->hw->id = AXI_DMAC_SG_UNUSED;
-		if (sg->schedule_when_free) {
-			sg->schedule_when_free = false;
-			start_next = true;
+	if (chan->hw_sg) {
+		if (active->cyclic) {
+			vchan_cyclic_callback(&active->vdesc);
+		} else {
+			list_del(&active->vdesc.node);
+			vchan_cookie_complete(&active->vdesc);
+			active = axi_dmac_active_desc(chan);
 		}
+	} else {
+		do {
+			sg = &active->sg[active->num_completed];
+			if (sg->hw->id == AXI_DMAC_SG_UNUSED) /* Not yet submitted */
+				break;
+			if (!(BIT(sg->hw->id) & completed_transfers))
+				break;
+			active->num_completed++;
+			sg->hw->id = AXI_DMAC_SG_UNUSED;
+			if (sg->schedule_when_free) {
+				sg->schedule_when_free = false;
+				start_next = true;
+			}
 
-		if (sg->partial_len)
-			axi_dmac_compute_residue(chan, active);
+			if (sg->partial_len)
+				axi_dmac_compute_residue(chan, active);
 
-		if (active->cyclic)
-			vchan_cyclic_callback(&active->vdesc);
+			if (active->cyclic)
+				vchan_cyclic_callback(&active->vdesc);
 
-		if (active->num_completed == active->num_sgs ||
-		    sg->partial_len) {
-			if (active->cyclic) {
-				active->num_completed = 0; /* wrap around */
-			} else {
-				list_del(&active->vdesc.node);
-				vchan_cookie_complete(&active->vdesc);
-				active = axi_dmac_active_desc(chan);
+			if (active->num_completed == active->num_sgs ||
+			    sg->partial_len) {
+				if (active->cyclic) {
+					active->num_completed = 0; /* wrap around */
+				} else {
+					list_del(&active->vdesc.node);
+					vchan_cookie_complete(&active->vdesc);
+					active = axi_dmac_active_desc(chan);
+				}
 			}
-		}
-	} while (active);
+		} while (active);
+	}
 
 	return start_next;
 }
@@ -481,8 +514,12 @@ static void axi_dmac_issue_pending(struct dma_chan *c)
 	struct axi_dmac_chan *chan = to_axi_dmac_chan(c);
 	struct axi_dmac *dmac = chan_to_axi_dmac(chan);
 	unsigned long flags;
+	u32 ctrl = AXI_DMAC_CTRL_ENABLE;
+
+	if (chan->hw_sg)
+		ctrl |= AXI_DMAC_CTRL_ENABLE_SG;
 
-	axi_dmac_write(dmac, AXI_DMAC_REG_CTRL, AXI_DMAC_CTRL_ENABLE);
+	axi_dmac_write(dmac, AXI_DMAC_REG_CTRL, ctrl);
 
 	spin_lock_irqsave(&chan->vchan.lock, flags);
 	if (vchan_issue_pending(&chan->vchan))
@@ -519,9 +556,13 @@ axi_dmac_alloc_desc(struct axi_dmac_chan *chan, unsigned int num_sgs)
 
 		hws[i].id = AXI_DMAC_SG_UNUSED;
 		hws[i].flags = 0;
+
+		/* Link hardware descriptors */
+		hws[i].next_sg_addr = hw_phys + (i + 1) * sizeof(*hws);
 	}
 
-	desc->num_sgs = num_sgs;
+	/* The last hardware descriptor will trigger an interrupt */
+	desc->sg[num_sgs - 1].hw->flags = AXI_DMAC_HW_FLAG_LAST | AXI_DMAC_HW_FLAG_IRQ;
 
 	return desc;
 }
@@ -781,6 +822,9 @@ static bool axi_dmac_regmap_rdwr(struct device *dev, unsigned int reg)
 	case AXI_DMAC_REG_DBG2:
 	case AXI_DMAC_REG_PARTIAL_XFER_LEN:
 	case AXI_DMAC_REG_PARTIAL_XFER_ID:
+	case AXI_DMAC_REG_CURRENT_SG_ID:
+	case AXI_DMAC_REG_SG_ADDRESS:
+	case AXI_DMAC_REG_SG_ADDRESS_HIGH:
 		return true;
 	default:
 		return false;
@@ -933,6 +977,10 @@ static int axi_dmac_detect_caps(struct axi_dmac *dmac, unsigned int version)
 	if (axi_dmac_read(dmac, AXI_DMAC_REG_FLAGS) == AXI_DMAC_FLAG_CYCLIC)
 		chan->hw_cyclic = true;
 
+	axi_dmac_write(dmac, AXI_DMAC_REG_SG_ADDRESS, 0xffffffff);
+	if (axi_dmac_read(dmac, AXI_DMAC_REG_SG_ADDRESS))
+		chan->hw_sg = true;
+
 	axi_dmac_write(dmac, AXI_DMAC_REG_Y_LENGTH, 1);
 	if (axi_dmac_read(dmac, AXI_DMAC_REG_Y_LENGTH) == 1)
 		chan->hw_2d = true;
@@ -1034,6 +1082,7 @@ static int axi_dmac_probe(struct platform_device *pdev)
 	dma_dev->dst_addr_widths = BIT(dmac->chan.dest_width);
 	dma_dev->directions = BIT(dmac->chan.direction);
 	dma_dev->residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
+	dma_dev->max_sg_burst = 31; /* 31 SGs maximum in one burst */
 	INIT_LIST_HEAD(&dma_dev->channels);
 
 	dmac->chan.vchan.desc_free = axi_dmac_desc_free;