Re: [PATCH 1/1] x86: Exclude E820_RESERVED regions and memory holes above 4 GB from direct mapping.

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

On 10/20/2011 03:10 PM, Jacob Shin wrote:
> On Thu, 2011-10-20 at 16:30 -0500, H. Peter Anvin wrote:
>> On 10/20/2011 02:28 PM, Andi Kleen wrote:
>>> Jacob Shin <jacob.shin@amd.com> writes:
>>>
>>>> On systems with very large memory (1 TB in our case), BIOS may report a
>>>> reserved region or a hole in the E820 map, even above the 4 GB range. Exclude
>>>> these from the direct mapping.
>>>
>>> This doesn't make much sense.  Holes above 4GB are completely legal.
>>>
>>> If you need to workaround a specific broken BIOS you would need a quirk
>>> only matching that system, with a suitable "BIOS is broken" message.
>>>
>>
>> The problem is that apparently right now we map those unconditionally
>> into the 1:1 map and mark them cacheable in PAT, which we *don't* for
>> the < 4 GiB map.
>>
>> This thus makes the behavior match < 4 GiB, which is the correct
>> behavior; this should be made clear in the patch description.
> 
> Will something like:
> 
> "On systems with very large memory (1 TB in our case), BIOS may report a
> reserved region or a hole in the E820 map, even above the 4 GB range.
> Avoid mapping them unconditionally into kernel 1:1 direct mapping as
> cacheable memory, as we also already do for the MMIO hole under 4 GB."
> 
> Work?
> 
> Otherwise, does the patch look acceptable?
> 

Drop the first half, and stop talking about "the MMIO hole" or anything
else with a definite article.

We're either doing this correctly for all holes or we have a bug.  If we
have a bug we should fix it in a general way, and I'm not convinced that
your patch is general enough.

I'm very bandwidth-constrained between kernel.org remediation and about
to leave for kernel summit, so I'm not sure how much detail I can look
into it right now.

	-hpa