[PATCH v4 2/4] spi: spi-mtk-nor: use dma_alloc_coherent() for bounce buffer

[Ikjoon Jang <ikjn@xxxxxxxxxxxx>]

Use dma_alloc_coherent() for bounce buffer instead of kmalloc() to
make sure the bounce buffer to be allocated within its DMAable range.

Additionally, add an internal helper need_bounce() function checking
whether op's data buffer is DMAable.

Reviewed-by: Chuanhong Guo <gch981213@xxxxxxxxx>
Signed-off-by: Ikjoon Jang <ikjn@xxxxxxxxxxxx>

---

Changes in v4:
- No changes since v3

Changes in v3:
- simplify function names
- restore back padding bytes to bounce buffer omitted in v2

 drivers/spi/spi-mtk-nor.c | 93 +++++++++++++++++++++------------------
 1 file changed, 51 insertions(+), 42 deletions(-)

diff --git a/drivers/spi/spi-mtk-nor.c b/drivers/spi/spi-mtk-nor.c
index 6e6ca2b8e6c8..cfb9a2450962 100644
--- a/drivers/spi/spi-mtk-nor.c
+++ b/drivers/spi/spi-mtk-nor.c
@@ -96,6 +96,7 @@ struct mtk_nor {
 	struct device *dev;
 	void __iomem *base;
 	u8 *buffer;
+	dma_addr_t buffer_dma;
 	struct clk *spi_clk;
 	struct clk *ctlr_clk;
 	unsigned int spi_freq;
@@ -144,6 +145,11 @@ static void mtk_nor_set_addr(struct mtk_nor *sp, const struct spi_mem_op *op)
 	}
 }
 
+static bool need_bounce(struct mtk_nor *sp, const struct spi_mem_op *op)
+{
+	return ((uintptr_t)op->data.buf.in & MTK_NOR_DMA_ALIGN_MASK);
+}
+
 static bool mtk_nor_match_read(const struct spi_mem_op *op)
 {
 	int dummy = 0;
@@ -169,6 +175,7 @@ static bool mtk_nor_match_read(const struct spi_mem_op *op)
 
 static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
 {
+	struct mtk_nor *sp = spi_controller_get_devdata(mem->spi->master);
 	size_t len;
 
 	if (!op->data.nbytes)
@@ -180,8 +187,7 @@ static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
 			if ((op->addr.val & MTK_NOR_DMA_ALIGN_MASK) ||
 			    (op->data.nbytes < MTK_NOR_DMA_ALIGN))
 				op->data.nbytes = 1;
-			else if (!((ulong)(op->data.buf.in) &
-				   MTK_NOR_DMA_ALIGN_MASK))
+			else if (!need_bounce(sp, op))
 				op->data.nbytes &= ~MTK_NOR_DMA_ALIGN_MASK;
 			else if (op->data.nbytes > MTK_NOR_BOUNCE_BUF_SIZE)
 				op->data.nbytes = MTK_NOR_BOUNCE_BUF_SIZE;
@@ -261,19 +267,12 @@ static void mtk_nor_setup_bus(struct mtk_nor *sp, const struct spi_mem_op *op)
 	mtk_nor_rmw(sp, MTK_NOR_REG_BUSCFG, reg, MTK_NOR_BUS_MODE_MASK);
 }
 
-static int mtk_nor_read_dma(struct mtk_nor *sp, u32 from, unsigned int length,
-			    u8 *buffer)
+static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
+			    dma_addr_t dma_addr)
 {
 	int ret = 0;
 	ulong delay;
 	u32 reg;
-	dma_addr_t dma_addr;
-
-	dma_addr = dma_map_single(sp->dev, buffer, length, DMA_FROM_DEVICE);
-	if (dma_mapping_error(sp->dev, dma_addr)) {
-		dev_err(sp->dev, "failed to map dma buffer.\n");
-		return -EINVAL;
-	}
 
 	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
 	writel(dma_addr, sp->base + MTK_NOR_REG_DMA_DADR);
@@ -298,30 +297,49 @@ static int mtk_nor_read_dma(struct mtk_nor *sp, u32 from, unsigned int length,
 					 (delay + 1) * 100);
 	}
 
-	dma_unmap_single(sp->dev, dma_addr, length, DMA_FROM_DEVICE);
 	if (ret < 0)
 		dev_err(sp->dev, "dma read timeout.\n");
 
 	return ret;
 }
 
-static int mtk_nor_read_bounce(struct mtk_nor *sp, u32 from,
-			       unsigned int length, u8 *buffer)
+static int mtk_nor_read_bounce(struct mtk_nor *sp, const struct spi_mem_op *op)
 {
 	unsigned int rdlen;
 	int ret;
 
-	if (length & MTK_NOR_DMA_ALIGN_MASK)
-		rdlen = (length + MTK_NOR_DMA_ALIGN) & ~MTK_NOR_DMA_ALIGN_MASK;
+	if (op->data.nbytes & MTK_NOR_DMA_ALIGN_MASK)
+		rdlen = (op->data.nbytes + MTK_NOR_DMA_ALIGN) & ~MTK_NOR_DMA_ALIGN_MASK;
 	else
-		rdlen = length;
+		rdlen = op->data.nbytes;
 
-	ret = mtk_nor_read_dma(sp, from, rdlen, sp->buffer);
-	if (ret)
-		return ret;
+	ret = mtk_nor_dma_exec(sp, op->addr.val, rdlen, sp->buffer_dma);
 
-	memcpy(buffer, sp->buffer, length);
-	return 0;
+	if (!ret)
+		memcpy(op->data.buf.in, sp->buffer, op->data.nbytes);
+
+	return ret;
+}
+
+static int mtk_nor_read_dma(struct mtk_nor *sp, const struct spi_mem_op *op)
+{
+	int ret;
+	dma_addr_t dma_addr;
+
+	if (need_bounce(sp, op))
+		return mtk_nor_read_bounce(sp, op);
+
+	dma_addr = dma_map_single(sp->dev, op->data.buf.in,
+				  op->data.nbytes, DMA_FROM_DEVICE);
+
+	if (dma_mapping_error(sp->dev, dma_addr))
+		return -EINVAL;
+
+	ret = mtk_nor_dma_exec(sp, op->addr.val, op->data.nbytes, dma_addr);
+
+	dma_unmap_single(sp->dev, dma_addr, op->data.nbytes, DMA_FROM_DEVICE);
+
+	return ret;
 }
 
 static int mtk_nor_read_pio(struct mtk_nor *sp, const struct spi_mem_op *op)
@@ -425,15 +443,8 @@ static int mtk_nor_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
 		if (op->data.nbytes == 1) {
 			mtk_nor_set_addr(sp, op);
 			return mtk_nor_read_pio(sp, op);
-		} else if (((ulong)(op->data.buf.in) &
-			    MTK_NOR_DMA_ALIGN_MASK)) {
-			return mtk_nor_read_bounce(sp, op->addr.val,
-						   op->data.nbytes,
-						   op->data.buf.in);
 		} else {
-			return mtk_nor_read_dma(sp, op->addr.val,
-						op->data.nbytes,
-						op->data.buf.in);
+			return mtk_nor_read_dma(sp, op);
 		}
 	}
 
@@ -588,7 +599,6 @@ static int mtk_nor_probe(struct platform_device *pdev)
 	struct spi_controller *ctlr;
 	struct mtk_nor *sp;
 	void __iomem *base;
-	u8 *buffer;
 	struct clk *spi_clk, *ctlr_clk;
 	int ret, irq;
 
@@ -604,16 +614,6 @@ static int mtk_nor_probe(struct platform_device *pdev)
 	if (IS_ERR(ctlr_clk))
 		return PTR_ERR(ctlr_clk);
 
-	buffer = devm_kmalloc(&pdev->dev,
-			      MTK_NOR_BOUNCE_BUF_SIZE + MTK_NOR_DMA_ALIGN,
-			      GFP_KERNEL);
-	if (!buffer)
-		return -ENOMEM;
-
-	if ((ulong)buffer & MTK_NOR_DMA_ALIGN_MASK)
-		buffer = (u8 *)(((ulong)buffer + MTK_NOR_DMA_ALIGN) &
-				~MTK_NOR_DMA_ALIGN_MASK);
-
 	ctlr = spi_alloc_master(&pdev->dev, sizeof(*sp));
 	if (!ctlr) {
 		dev_err(&pdev->dev, "failed to allocate spi controller\n");
@@ -633,13 +633,22 @@ static int mtk_nor_probe(struct platform_device *pdev)
 
 	sp = spi_controller_get_devdata(ctlr);
 	sp->base = base;
-	sp->buffer = buffer;
 	sp->has_irq = false;
 	sp->wbuf_en = false;
 	sp->ctlr = ctlr;
 	sp->dev = &pdev->dev;
 	sp->spi_clk = spi_clk;
 	sp->ctlr_clk = ctlr_clk;
+	sp->buffer = dmam_alloc_coherent(&pdev->dev,
+				MTK_NOR_BOUNCE_BUF_SIZE + MTK_NOR_DMA_ALIGN,
+				&sp->buffer_dma, GFP_KERNEL);
+	if (!sp->buffer)
+		return -ENOMEM;
+
+	if ((uintptr_t)sp->buffer & MTK_NOR_DMA_ALIGN_MASK) {
+		dev_err(sp->dev, "misaligned allocation of internal buffer.\n");
+		return -ENOMEM;
+	}
 
 	irq = platform_get_irq_optional(pdev, 0);
 	if (irq < 0) {
-- 
2.28.0.709.gb0816b6eb0-goog