On Monday, May 18, 2015 05:38:17 PM Suravee Suthikulanit wrote:

Hi Rafael,

On 5/15/2015 6:53 PM, Rafael J. Wysocki wrote:

quoted

On Friday, May 15, 2015 04:23:09 PM Suravee Suthikulpanit wrote:

quoted

[...]

diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
index 4bf7559..f6bc438 100644
--- a/drivers/acpi/acpi_platform.c
+++ b/drivers/acpi/acpi_platform.c

@@ -103,14 +103,18 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
  	pdevinfo.res = resources;
  	pdevinfo.num_res = count;
  	pdevinfo.fwnode = acpi_fwnode_handle(adev);
-	pdevinfo.dma_mask = DMA_BIT_MASK(32);
+	pdevinfo.dma_mask = acpi_dma_is_supported(adev) ? DMA_BIT_MASK(32) : 0;
  	pdev = platform_device_register_full(&pdevinfo);
-	if (IS_ERR(pdev))
+	if (IS_ERR(pdev)) {
  		dev_err(&adev->dev, "platform device creation failed: %ld\n",
  			PTR_ERR(pdev));
-	else
+	} else {
+		if (acpi_dma_is_supported(adev))
+			arch_setup_dma_ops(&pdev->dev, 0, 0, NULL,
+					   acpi_dma_is_coherent(adev));

Shouldn't we generally do that in acpi_bind_one() for all bus types
that don't have specific handling rather than here?

I think that would also work, and makes sense. However, I'm not sure if 
this would help in the case when we are creating PCI end-point devices, 
since the CCA is specified at the host bridge node, and there is no ACPI 
companion for the end-point devices. It seems that patch 3/6 of this 
series is still needed.

Yes, PCI needs its own handling, but there are multiple bus types that
don't (SPI, I2C etc) in addition to the platform bus type.


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.