Re: [RFC 04/42] drivers/dma: don't check resource with devm_ioremap_resource

[Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org>]

On 05/10/2013 02:16 AM, Wolfram Sang wrote:
> devm_ioremap_resource does sanity checks on the given resource. No need to
> duplicate this in the driver.

> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c

>  	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> -	if (!res) {
> -		dev_err(&pdev->dev, "No mem resource for DMA\n");
> -		return -EINVAL;
> -	}
> -
>  	tdma->base_addr = devm_ioremap_resource(&pdev->dev, res);

One issue here is that it's not obvious just from reading the code
that's left behind that the "missing" error-checking of the
platform_get_resource() return value is OK because
devm_ioremap_resource() will check it "for us". Everyone now has to
mentally maintain a list of exceptions where it's OK not to error-check.
A similar situation exists for e.g. kzalloc already spewing an error
when allocations fail, and so drivers don't need to print a diagnostic
in that case, but it's helpful if they do in most other cases, but
that's another issue.

Would it be better to introduce a new devm_ioremap_pdev_resource(pdev,
index) that replaced both those two API calls with a single one. That
way, only the author/reader of the new devm_ioremap_pdev_resource()
would have to remember that caveat, and a single comment could be added
so people unfamiliar with the code remember, without duplicating the
comment in every driver.