On 09/28/2017 10:13 AM, Vinod Koul wrote:

On Fri, Aug 25, 2017 at 04:31:04PM +0200, Pierre-Yves MORDRET wrote:

quoted

+static int stm32_mdma_probe(struct platform_device *pdev)
+{

[snip]

quoted

+	dmadev = devm_kzalloc(&pdev->dev, sizeof(*dmadev) + sizeof(u32) * count,
+			      GFP_KERNEL);
+	if (!dmadev)
+		return -ENOMEM;
+
+	dmadev->nr_channels = nr_channels;
+	dmadev->nr_requests = nr_requests;
+	device_property_read_u32_array(&pdev->dev, "st,ahb-addr-masks",
+				       dmadev->ahb_addr_masks,
+				       count);
+	dmadev->nr_ahb_addr_masks = count;
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	dmadev->base = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(dmadev->base))
+		return PTR_ERR(dmadev->base);
+
+	dmadev->clk = devm_clk_get(&pdev->dev, NULL);
+	if (IS_ERR(dmadev->clk)) {
+		dev_err(&pdev->dev, "Error: Missing controller clock\n");
+		return PTR_ERR(dmadev->clk);
+	}

I am assuming if clk is not availble at this point of time you return clk
error which maybe defer probe, right?

Yes. So I will add an "if" statement as I did for DMAMUX
if (ret == -EPROBE_DEFER)
   dev_info .... (instead of dev_err)

quoted

+static struct platform_driver stm32_mdma_driver = {
+	.driver = {
+		.name = "stm32-mdma",
+		.of_match_table = stm32_mdma_of_match,
+	},
+};
+
+static int __init stm32_mdma_init(void)
+{
+	return platform_driver_probe(&stm32_mdma_driver, stm32_mdma_probe);

and you register platform_driver_probe() which per
drivers/base/platform.c:760 is incompatible with defer probing.
So this need a fix

Will change to platform_driver_register

Thanks and regards