Re: [PATCH] TESTCASE: of: OOPS when disabling node via OF_DYNAMIC

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Grant,

On Sunday 07 June 2015 12:16:32 Grant Likely wrote:
> On Wed, 22 Apr 2015 15:30:28 +0300, Pantelis Antoniou wrote:
> > > On Apr 14, 2015, at 16:27 , Wolfram Sang <wsa@the-dreams.de> wrote:
> > > 
> > > Hi Pantelis,
> > > 
> > > thanks for your prompt reply. Unfortunately, I had to wait until I could
> > > access the test system again.
> > 
> > [snip]
> > 
> > Sorry for the non-prompt reply; but just for kicks, can you try the
> > attached patch?
> > 
> > I have a hunch this might be the problem.
> > 
> > Regards
> > 
> > — Pantelis
> 
> I played around with this some today. If I'm reading it correctly, the
> following patch reproduces the same problem:
> 
> (continued below patch)
> ---
> diff --git a/drivers/of/unittest.c b/drivers/of/unittest.c
> index 18016341d5a9..0a27b38c3041 100644
> --- a/drivers/of/unittest.c
> +++ b/drivers/of/unittest.c
> @@ -753,6 +753,11 @@ static void __init of_unittest_match_node(void)
>  	}
>  }
> 
> +static struct resource test_bus_res = {
> +	.start = 0xfffffff8,
> +	.end = 0xfffffff9,
> +	.flags = IORESOURCE_MEM,
> +};
>  static const struct platform_device_info test_bus_info = {
>  	.name = "unittest-bus",
>  };
> @@ -795,6 +800,7 @@ static void __init of_unittest_platform_populate(void)
>  	if (rc)
>  		return;
>  	test_bus->dev.of_node = np;
> +	platform_device_add_resources(test_bus, &test_bus_res, 1);
> 
>  	of_platform_populate(np, match, NULL, &test_bus->dev);
>  	for_each_child_of_node(np, child) {
> 
> ---
> 
> I think the fixup patch boils down to the following. It's hard to tell
> because it combines refactoring with the bug fix. Do I have it correct?
> If so, I'd rather commit the simple fix which can be committed now for
> v4.1, and the refactoring can be pushed for v4.2

The patch below looks equivalent to Pantelis' patch if we remove the 
refactoring, but I don't immediately see how r->parent can be NULL in 
platform_device_del() if type is equal to IORESOURCE_MEM or IORESOURCE_IO, as 
platform_device_add() will have called insert_resource() in those cases, which 
should set the resource parent pointer. I must be missing something as the 
patch fixes Wolfram's issue, so an explanation in the commit message would be 
welcome.

> ---
> 
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index ebf034b97278..b3042e6ee3ef 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -377,7 +377,7 @@ int platform_device_add(struct platform_device *pdev)
>  		struct resource *r = &pdev->resource[i];
>  		unsigned long type = resource_type(r);
> 
> -		if (type == IORESOURCE_MEM || type == IORESOURCE_IO)
> +		if (r->parent)
>  			release_resource(r);
>  	}
> 
> @@ -410,7 +410,7 @@ void platform_device_del(struct platform_device *pdev)
>  			struct resource *r = &pdev->resource[i];
>  			unsigned long type = resource_type(r);
> 
> -			if (type == IORESOURCE_MEM || type == IORESOURCE_IO)
> +			if (r->parent)
>  				release_resource(r);
>  		}
>  	}

-- 
Regards,

Laurent Pinchart