Re: Radical idea

Re: Radical idea

[Christoph Lameter]

Sorry about the wrong email address. Bouncing does not do proper outgoing 
address translation.

> Andi wrote:
> > x86-64 can have multiple zones in node > 0 (e.g. node 1 can have both
> > DMA32 and NORMAL) 

>In this case, Christoph, would your radical idea preserve user visible
>node numbers?  In general, the kernels numbering of nodes (as well as
>its numbering of cpus) is exposed to user space in various ways.  What's
>exposed should not change.

It would just add new node numbers for containers and dma zones outside 
of the physical range.

And yes it would only work the DMA32 problems mentioned by Andi could be
addressed. Do we really need DMA32 in modern systems with IOMMUs? Isnt 
this a transitionary problem that will go away?

So lets say we have one of those systems without IOMMU. Then we only have 
a problem for a class of NUMA systems that have:

1. Memory beyond 4GB

and

2. Per node memory less than 4GB. Otherwise DMA32 is only on node 0.

Isnt this a fairly small group of systems?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: Radical idea

[Paul Jackson]

Christoph wrote:
> It would just add new node numbers for containers and dma zones outside 
> of the physical range.

Good.

> And yes it would only work the DMA32 problems mentioned by Andi could be
> addressed.

Good luck ;).

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: Radical idea

[Andi Kleen]

> 
> And yes it would only work the DMA32 problems mentioned by Andi could be
> addressed. Do we really need DMA32 in modern systems with IOMMUs?

We do.

In fact we still need DMA-without-32 on most systems (after all most users
still want to use their floppy occasionally and it is needed for a few other
devices too)

> Isnt this a transitionary problem that will go away?

Not any time soon.

> So lets say we have one of those systems without IOMMU. Then we only have 
> a problem for a class of NUMA systems that have:
> 
> 1. Memory beyond 4GB
> 
> and
> 
> 2. Per node memory less than 4GB. Otherwise DMA32 is only on node 0.
> 
> Isnt this a fairly small group of systems?

I don't think so. e.g. a lot of quad opteron configurations come with 1 or 2GB per
socket (= node) 

Anyways, even if it was uncommon we couldn't just break it. So i'm not sure what
the point of your "popularity contest" is?

-Andi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>