2005-11-11 16:06:31

by Anton Blanchard

[permalink] [raw]
Subject: [PATCH] Allow flatmem to be disabled when only sparsemem is implemented


Hi Andy,

What do you think about this?

Anton

--

On architectures that implement sparsemem but not discontigmem we want
to be able to hide the flatmem option in some cases. On ppc64 for
example, when we select NUMA we must not select flatmem.

Signed-off-by: Anton Blanchard <[email protected]>
---

Index: build/mm/Kconfig
===================================================================
--- build.orig/mm/Kconfig 2005-11-11 11:31:38.000000000 +1100
+++ build/mm/Kconfig 2005-11-11 11:41:55.000000000 +1100
@@ -11,7 +11,7 @@

config FLATMEM_MANUAL
bool "Flat Memory"
- depends on !ARCH_DISCONTIGMEM_ENABLE || ARCH_FLATMEM_ENABLE
+ depends on !(ARCH_DISCONTIGMEM_ENABLE || ARCH_SPARSEMEM_ENABLE) || ARCH_FLATMEM_ENABLE
help
This option allows you to change some of the ways that
Linux manages its memory internally. Most users will


2005-11-11 17:52:15

by Andy Whitcroft

[permalink] [raw]
Subject: Re: [PATCH] Allow flatmem to be disabled when only sparsemem is implemented

Anton Blanchard wrote:

> On architectures that implement sparsemem but not discontigmem we want
> to be able to hide the flatmem option in some cases. On ppc64 for
> example, when we select NUMA we must not select flatmem.
>
> Signed-off-by: Anton Blanchard <[email protected]>

First reaction is that this is very reasonable. I can see why you need
to do this as you don't have DISCONTIGMEM. I will just go check the
major architectures and make sure they arn't relying on being able to
enable SPARSEMEM and getting FLATMEM too behaviour. I don't think they
can be as they all have DISCONTIGMEM and so should be insulated.

-apw

2005-11-11 19:41:57

by Andy Whitcroft

[permalink] [raw]
Subject: Re: [PATCH] Allow flatmem to be disabled when only sparsemem is implemented

Andy Whitcroft wrote:
> Anton Blanchard wrote:
>
>>On architectures that implement sparsemem but not discontigmem we want
>>to be able to hide the flatmem option in some cases. On ppc64 for
>>example, when we select NUMA we must not select flatmem.
>
> First reaction is that this is very reasonable. I can see why you need
> to do this as you don't have DISCONTIGMEM. I will just go check the
> major architectures and make sure they arn't relying on being able to
> enable SPARSEMEM and getting FLATMEM too behaviour. I don't think they
> can be as they all have DISCONTIGMEM and so should be insulated.

Ok. I've reviewed the usage of the memory model selectors in the
architectures in 2.6.14-mm2. It appears that only i386 is affected by
this change, the others that use the selector have explicit enablement
of FLATMEM. This patch will interact badly with the current code to
enable SPARSEMEM on non-numa systems. However, this code is under
review at this moment, and the proposed replacement (Message-ID:
<[email protected]>) is compatible with this change.

In short as long as they go in together this change looks good.

Magnus, any feedback on the replacement SPARSEMEM enabler on non-NUMA
i386 systems??

-apw

2005-11-14 00:51:34

by Magnus Damm

[permalink] [raw]
Subject: Re: [PATCH] Allow flatmem to be disabled when only sparsemem is implemented

On Fri, 2005-11-11 at 19:41 +0000, Andy Whitcroft wrote:
> Andy Whitcroft wrote:
> > Anton Blanchard wrote:
> >
> >>On architectures that implement sparsemem but not discontigmem we want
> >>to be able to hide the flatmem option in some cases. On ppc64 for
> >>example, when we select NUMA we must not select flatmem.
> >
> > First reaction is that this is very reasonable. I can see why you need
> > to do this as you don't have DISCONTIGMEM. I will just go check the
> > major architectures and make sure they arn't relying on being able to
> > enable SPARSEMEM and getting FLATMEM too behaviour. I don't think they
> > can be as they all have DISCONTIGMEM and so should be insulated.
>
> Ok. I've reviewed the usage of the memory model selectors in the
> architectures in 2.6.14-mm2. It appears that only i386 is affected by
> this change, the others that use the selector have explicit enablement
> of FLATMEM. This patch will interact badly with the current code to
> enable SPARSEMEM on non-numa systems. However, this code is under
> review at this moment, and the proposed replacement (Message-ID:
> <[email protected]>) is compatible with this change.
>
> In short as long as they go in together this change looks good.
>
> Magnus, any feedback on the replacement SPARSEMEM enabler on non-NUMA
> i386 systems??

I tested your patch in QEMU on top of 2.6.15-rc1-git1 and it seems to
work well. Many thanks!

/ magnus