Re: [patch 2/4] add x86-64 Kconfig options for sparsemem

Re: [patch 2/4] add x86-64 Kconfig options for sparsemem

[Matt Tolentino]

>On Wed, May 18, 2005 at 08:24:41AM -0700, Matt Tolentino wrote:
>> 
>> Add the requisite arch specific Kconfig options to enable 
>> the use of the sparsemem implementation for NUMA kernels
>> on x86-64.
>
>How much did you test sparsemem on x86-64 NUMA ? 
>
>There are various cases that probably need to be checked,
>AMD with SRAT, AMD without SRAT, AMD with more than 4GB RAM, 
>Summit(?), NUMA EMULATION etc.
>
>If all that works I would have no problem with removing the
>old code.

As my disclaimer said, this has only been tested using
the NUMA EMULATION config option.  That's a big part of
the reason for sending this out  - to get further testing 
on real x86-64 NUMA systems, but without breaking the
current discontigmem code.  

I expect to be able to test this on at least one AMD system
at the local University systems lab, but haven't had a
chance to do so yet.

matt

--
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:"aart@kvack.org"> aart@kvack.org </a>

Re: [patch 2/4] add x86-64 Kconfig options for sparsemem

[Andi Kleen]

On Wed, May 18, 2005 at 09:43:48AM -0700, Matt Tolentino wrote:
> >From: Andi Kleen <ak@muc.de>
> >On Wed, May 18, 2005 at 08:24:41AM -0700, Matt Tolentino wrote:
> >> 
> >> Add the requisite arch specific Kconfig options to enable 
> >> the use of the sparsemem implementation for NUMA kernels
> >> on x86-64.
> >
> >How much did you test sparsemem on x86-64 NUMA ? 
> >
> >There are various cases that probably need to be checked,
> >AMD with SRAT, AMD without SRAT, AMD with more than 4GB RAM, 
> >Summit(?), NUMA EMULATION etc.
> >
> >If all that works I would have no problem with removing the
> >old code.
> 
> As my disclaimer said, this has only been tested using
> the NUMA EMULATION config option.  That's a big part of
> the reason for sending this out  - to get further testing 
> on real x86-64 NUMA systems, but without breaking the
> current discontigmem code.  

Hmm, I would have assumed IBM tested it, since Dave Hansen signed off - 
they have a range of Opteron machines.   If not I can test it
on a few boxes later.

A single box is not enough, there are various special cases. 
e.g. one area I've been fighting with is that
with SRAT and 3+GB memory the nodes don't span the PCI memory
hole anymore, and when there is a virt_to_page() or similar for these
addresses things go wrong because they do a uninitialized hash
table lookup. If it's not that hard and doesn't cause code bloat
I would recommend to harden sparsemem against this case, at least for 
upto 4GB.

-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:"aart@kvack.org"> aart@kvack.org </a>

Re: [patch 2/4] add x86-64 Kconfig options for sparsemem

[Dave Hansen]

On Wed, 2005-05-18 at 18:53 +0200, Andi Kleen wrote:
> Hmm, I would have assumed IBM tested it, since Dave Hansen signed off - 
> they have a range of Opteron machines.   If not I can test it
> on a few boxes later.

I actually don't personally have any access to Opteron machines.  But, I
know Keith Mannthey has been testing it all along on his various x86_64
machines.  I'll certainly make sure we get another run on all of those
once it goes into -mm.

-- Dave

--
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:"aart@kvack.org"> aart@kvack.org </a>

Re: [patch 2/4] add x86-64 Kconfig options for sparsemem

[Andi Kleen]

On Wed, May 18, 2005 at 08:24:41AM -0700, Matt Tolentino wrote:
> 
> Add the requisite arch specific Kconfig options to enable 
> the use of the sparsemem implementation for NUMA kernels
> on x86-64.

How much did you test sparsemem on x86-64 NUMA ? 

There are various cases that probably need to be checked,
AMD with SRAT, AMD without SRAT, AMD with more than 4GB RAM, 
Summit(?), NUMA EMULATION etc.

If all that works I would have no problem with removing the
old code.

-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:"aart@kvack.org"> aart@kvack.org </a>

[patch 2/4] add x86-64 Kconfig options for sparsemem

[Matt Tolentino]

Add the requisite arch specific Kconfig options to enable 
the use of the sparsemem implementation for NUMA kernels
on x86-64.

Signed-off-by: Matt Tolentino <matthew.e.tolentino@intel.com>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
---

 arch/x86_64/Kconfig |   19 +++++++++++++++++++
 1 files changed, 19 insertions(+)

diff -urNp linux-2.6.12-rc4-mm2/arch/x86_64/Kconfig linux-2.6.12-rc4-mm2-m/arch/x86_64/Kconfig
--- linux-2.6.12-rc4-mm2/arch/x86_64/Kconfig	2005-05-18 07:38:20.000000000 -0400
+++ linux-2.6.12-rc4-mm2-m/arch/x86_64/Kconfig	2005-05-18 07:44:42.000000000 -0400
@@ -274,8 +274,27 @@ config NUMA
        bool
        default n
 
+config ARCH_DISCONTIGMEM_ENABLE
+	def_bool y
+	depends on NUMA
+
+config ARCH_DISCONTIGMEM_DEFAULT
+	def_bool y
+	depends on NUMA
+
+config ARCH_SPARSEMEM_ENABLE
+	def_bool y
+	depends on NUMA
+
+config ARCH_FLATMEM_ENABLE
+	def_bool y
+	depends on !NUMA
+
 source "mm/Kconfig"
 
+config HAVE_ARCH_EARLY_PFN_TO_NID
+	def_bool y
+
 config HAVE_DEC_LOCK
 	bool
 	depends on SMP
--
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:"aart@kvack.org"> aart@kvack.org </a>