Re: pcpu allocator on large NUMA machines

[Tejun Heo]

Hello, Michael.

On Tue, Jul 25, 2017 at 11:26:03AM +1000, Michael Ellerman wrote:
> I don't think we want to stop using embed first chunk unless we have to.
> 
> We have code that accesses percpu variables in real mode (with the MMU
> off), and that wouldn't work easily if the first chunk wasn't in the
> linear mapping. So it's not just an optimisation for us.
> 
> We can fairly easily make the vmalloc space 56T, and I'm working on a
> patch to make it ~500T on newer machines.

Yeah, the only constraint is the size of vmalloc area in relation to
the maximum spread across NUMA regions.  If the vmalloc space can be
made bigger, that'd be the best option.  As the area percpu allocator
actually uses is very small comparatively, it doesn't have to be a lot
larger either.

Thanks.

-- 
tejun