Re: WARNING: CPU: 2 PID: 1 at arch/x86/mm/ioremap.c:171 __ioremap_caller+0x290/0x2fa()

["H. Peter Anvin" <hpa@zytor.com>]

On 07/15/2014 04:40 PM, Yinghai Lu wrote:
>>
>> One of the reasons for iomem_resource is so we don't hand out the same
>> address space to two different devices.  We *could* do that by keeping
>> track of the union of all devices and reserved areas that we know
>> about.
>>
>> But the current resource code is more strict: it enforces a hierarchy.
>> For example, in this case, it rejects the 00:00 PNP resource because
>> it is larger than the e820 entry.  The problem with rejecting it is
>> that we might hand out [mem 0xfed14000-0xfed17fff] to another device
>> even though PNP told us that it's in use.
>>
>> I'm about to head out for a few weeks of vacation, so I won't be able
>> to do anything with this.
> 
> In that case, we could reserve the whole MCH range in e820 from
> trim_snb_memory() instead.
> 
> HPA, what is your idea about it?
> 
> Yinghai
> 

We could quirk it, but we would have to make bloody darn sure that we
don't break any systems because of unusual configuration and so on.

I agree that we need to treat fixed resources as equivalent to reserved.
 This is also a BIOS bug (it should reserve the whole region), but that
happens far too frequently.  I don't know if we have any way to do that
without massive surgery to the current code, though.

	-hpa