Re: [PATCH v3] mm/hugetlb: split hugetlb_cma in nodes with memory

[Mike Kravetz]

On 7/27/20 7:37 AM, Aneesh Kumar K.V wrote:
> There is variant of this which is pseries powerpc where there is
> hypervisor assistance w.r.t allocating gigantic pages. See the ppc64
> enablement patch 
> 
> https://lore.kernel.org/linuxppc-dev/20200713150749.25245-1-aneesh.kumar@xxxxxxxxxxxxx/
> 

Thanks Aneesh,  I missed the powerpc support.

I knew about the powerpc hypervisor assistance which caused me to rethink
my original idea that all this could be arch independent.  My next idea was
that architectures would only need to provide a constant something like:

#define HUGETLB_CMA_ORDER (PUD_SHIFT - PAGE_SHIFT)

However, it looks like this can not be a compile time constant on powerpc.

Moving more of the CMA support to arch independent code has moved down on
my priority list.  So, it it likely not to get much work in the immediate
future.

Just curious, can you have multiple gigantic page sizes supported at one
time (one system instance) on powerpc?
-- 
Mike Kravetz