Re: [PATCH 1/3] x86, mm: Include the entire kernel memory map in trampoline_pgd

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

Actually, no, because we may need to double-map in the low address space per discussion at LPC.

Jan Beulich <jbeulich@suse.com> wrote:

>>>> "H. Peter Anvin" <hpa@zytor.com> 10/05/12 6:28 PM >>>
>>On 10/04/2012 11:39 PM, Jan Beulich wrote:
>>>>
>>>> We should have the check, but at least for Linux support we require
>>>> P <= V-2.
>>>
>>> Not really imo - P <= V - 1 should be sufficient here, as all that
>is
>>> necessary is that the result represents a 1:1 mapping. Specifically,
>>> there's no constraint to the virtual space limitation of the direct
>>> mapping of RAM.
>>>
>>
>>The P <= V-2 limitation doesn't come from this, it comes from the fact
>
>>that we need to have the regular kernel 1:1 map and still have space
>for 
>>the kernel, vmalloc, ioremap and so on in the kernel part of the
>address 
>>space; in theory it *could* be some fraction between 1 and 2, but
>since 
>>hardware doesn't do fractional bits very well the above is what we
>have 
>>been telling the hardware folks. :)
>
>I understand all that, but here we're talking about a true 1:1 mapping
>(i.e. in the lower half of the virtual address space), as opposed to
>the
>direct mapping (in the upper half). We can obviously use all of the
>lower
>half for this purpose.
>
>Jan

-- 
Sent from my mobile phone. Please excuse brevity and lack of formatting.