[PATCH v2 1/2] arm64: vmemmap: use virtual projection of linear region

[Mark Langsdorf]

On 03/08/2016 04:31 AM, Ard Biesheuvel wrote:
> On 8 March 2016 at 09:15, Ard Biesheuvel <ard.biesheuvel at linaro.org> wrote:
>>
>>
>>> On 8 mrt. 2016, at 08:07, David Daney <ddaney.cavm at gmail.com> wrote:
>>>
>>>> On 02/26/2016 08:57 AM, Ard Biesheuvel wrote:
>>>> Commit dd006da21646 ("arm64: mm: increase VA range of identity map") made
>>>> some changes to the memory mapping code to allow physical memory to reside
>>>> at an offset that exceeds the size of the virtual mapping.
>>>>
>>>> However, since the size of the vmemmap area is proportional to the size of
>>>> the VA area, but it is populated relative to the physical space, we may
>>>> end up with the struct page array being mapped outside of the vmemmap
>>>> region. For instance, on my Seattle A0 box, I can see the following output
>>>> in the dmesg log.
>>>>
>>>>     vmemmap : 0xffffffbdc0000000 - 0xffffffbfc0000000   (     8 GB maximum)
>>>>               0xffffffbfc0000000 - 0xffffffbfd0000000   (   256 MB actual)
>>>>
>>>> We can fix this by deciding that the vmemmap region is not a projection of
>>>> the physical space, but of the virtual space above PAGE_OFFSET, i.e., the
>>>> linear region. This way, we are guaranteed that the vmemmap region is of
>>>> sufficient size, and we can even reduce the size by half.
>>>>
>>>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel at linaro.org>
>>>
>>> I see this commit now in Linus' kernel.org tree in v4.5-rc7.
>>>
>>> FYI:  I am seeing a crash that goes away when I revert this.  My kernel has some other modifications (our NUMA patches) so I haven't yet fully tracked this down on an unmodified kernel, but this is what I am getting:
>>>
>>
>
> I managed to reproduce and diagnose this. The problem is that vmemmap
> is no longer zone aligned, which causes trouble in the zone based
> rounding that occurs in memory_present. The below patch fixes this by
> rounding down the subtracted offset. Since this implies that the
> region could stick off the other end, it also reverts the halving of
> the region size.

This fixes the bug on my Seattle B0 system.

Tested-by: Mark Langsdorf <mlangsdo at redhat.com>

--Mark Langsdorf