Nornal memory CPU copy with cache invalidate is more efficient
than uncache memory copy.

Signed-off-by: Jon Lin <jon.lin@rock-chips.com>
---

 drivers/spi/spi-rockchip-sfc.c | 47 ++++++++++++++++++----------------
 1 file changed, 25 insertions(+), 22 deletions(-)

diff --git a/drivers/spi/spi-rockchip-sfc.c b/drivers/spi/spi-rockchip-sfc.c
index ca4f20283d23..36509bcccc27 100644
--- a/drivers/spi/spi-rockchip-sfc.c
+++ b/drivers/spi/spi-rockchip-sfc.c
@@ -156,10 +156,9 @@
 
 #define SFC_MAX_CHIPSELECT_NUM		2
 
-/* The SFC can transfer max 16KB - 1 at one time
- * we set it to 15.5KB here for alignment.
- */
 #define SFC_MAX_IOSIZE_VER3		(512 * 31)
+/* Although up to 4GB, 64KB is enough with less mem reserved */
+#define SFC_MAX_IOSIZE_VER4		(0x10000U)
 
 /* DMA is only enabled for large data transmission */
 #define SFC_DMA_TRANS_THRETHOLD		(0x40)
@@ -457,8 +456,10 @@ static int rockchip_sfc_xfer_data_dma(struct rockchip_sfc *sfc,
 
 	dev_dbg(sfc->dev, "sfc xfer_dma len=%x\n", len);
 
-	if (op->data.dir == SPI_MEM_DATA_OUT)
+	if (op->data.dir == SPI_MEM_DATA_OUT) {
 		memcpy(sfc->buffer, op->data.buf.out, len);
+		dma_sync_single_for_device(sfc->dev, sfc->dma_buffer, len, DMA_TO_DEVICE);
+	}
 
 	ret = rockchip_sfc_fifo_transfer_dma(sfc, sfc->dma_buffer, len);
 	if (!wait_for_completion_timeout(&sfc->cp, msecs_to_jiffies(2000))) {
@@ -466,8 +467,11 @@ static int rockchip_sfc_xfer_data_dma(struct rockchip_sfc *sfc,
 		ret = -ETIMEDOUT;
 	}
 	rockchip_sfc_irq_mask(sfc, SFC_IMR_DMA);
-	if (op->data.dir == SPI_MEM_DATA_IN)
+
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		dma_sync_single_for_cpu(sfc->dev, sfc->dma_buffer, len, DMA_FROM_DEVICE);
 		memcpy(op->data.buf.in, sfc->buffer, len);
+	}
 
 	return ret;
 }
@@ -633,19 +637,6 @@ static int rockchip_sfc_probe(struct platform_device *pdev)
 
 	sfc->use_dma = !of_property_read_bool(sfc->dev->of_node, "rockchip,sfc-no-dma");
 
-	if (sfc->use_dma) {
-		ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
-		if (ret) {
-			dev_warn(dev, "Unable to set dma mask\n");
-			return ret;
-		}
-
-		sfc->buffer = dmam_alloc_coherent(dev, SFC_MAX_IOSIZE_VER3,
-						  &sfc->dma_buffer, GFP_KERNEL);
-		if (!sfc->buffer)
-			return -ENOMEM;
-	}
-
 	ret = clk_prepare_enable(sfc->hclk);
 	if (ret) {
 		dev_err(&pdev->dev, "Failed to enable ahb clk\n");
@@ -676,8 +667,8 @@ static int rockchip_sfc_probe(struct platform_device *pdev)
 	if (ret)
 		goto err_irq;
 
-	sfc->max_iosize = rockchip_sfc_get_max_iosize(sfc);
 	sfc->version = rockchip_sfc_get_version(sfc);
+	sfc->max_iosize = rockchip_sfc_get_max_iosize(sfc);
 
 	pm_runtime_set_autosuspend_delay(dev, ROCKCHIP_AUTOSUSPEND_DELAY);
 	pm_runtime_use_autosuspend(dev);
@@ -685,16 +676,27 @@ static int rockchip_sfc_probe(struct platform_device *pdev)
 	pm_runtime_enable(dev);
 	pm_runtime_get_noresume(dev);
 
+	if (sfc->use_dma) {
+		sfc->buffer = (u8 *)__get_free_pages(GFP_KERNEL | GFP_DMA32,
+						     get_order(sfc->max_iosize));
+		if (!sfc->buffer) {
+			ret = -ENOMEM;
+			goto err_dma;
+		}
+		sfc->dma_buffer = virt_to_phys(sfc->buffer);
+	}
+
 	ret = devm_spi_register_controller(dev, host);
 	if (ret)
-		goto err_pm_runtime_free;
+		goto err_register;
 
 	pm_runtime_mark_last_busy(dev);
 	pm_runtime_put_autosuspend(dev);
 
 	return 0;
-
-err_pm_runtime_free:
+err_register:
+	free_pages((unsigned long)sfc->buffer, get_order(sfc->max_iosize));
+err_dma:
 	pm_runtime_get_sync(dev);
 	pm_runtime_put_noidle(dev);
 	pm_runtime_disable(dev);
@@ -714,6 +716,7 @@ static void rockchip_sfc_remove(struct platform_device *pdev)
 	struct spi_controller *host = sfc->host;
 
 	spi_unregister_controller(host);
+	free_pages((unsigned long)sfc->buffer, get_order(sfc->max_iosize));
 
 	clk_disable_unprepare(sfc->clk);
 	clk_disable_unprepare(sfc->hclk);
-- 
2.34.1

On Thu, 19 Dec 2024 09:05:57 +0800, Jon Lin wrote:
> Nornal memory CPU copy with cache invalidate is more efficient
> than uncache memory copy.
> 
> 

Applied to

   https://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git for-next

Thanks!

[1/1] spi: rockchip-sfc: Using normal memory for dma
      commit: b69386fcbc6066fb4885667743ab4d4967d561b8

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark