[PATCH] - Increase max physical memory size of x86_64

[PATCH] - Increase max physical memory size of x86_64

[Jack Steiner]

Increase the maximum physical address size of x86_64 system
to 44-bits. This is in preparation for future chips that
support larger physical memory sizes.

	Signed-off-by: Jack Steiner <steiner@sgi.com>

---
 include/asm-x86/sparsemem.h |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: linux/include/asm-x86/sparsemem.h
===================================================================
--- linux.orig/include/asm-x86/sparsemem.h	2008-03-10 00:22:27.000000000 -0500
+++ linux/include/asm-x86/sparsemem.h	2008-03-11 14:46:29.000000000 -0500
@@ -26,8 +26,8 @@
 # endif
 #else /* CONFIG_X86_32 */
 # define SECTION_SIZE_BITS	27 /* matt - 128 is convenient right now */
-# define MAX_PHYSADDR_BITS	40
-# define MAX_PHYSMEM_BITS	40
+# define MAX_PHYSADDR_BITS	44
+# define MAX_PHYSMEM_BITS	44
 #endif
 
 #endif /* CONFIG_SPARSEMEM */

--
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: [PATCH] - Increase max physical memory size of x86_64

[Ingo Molnar]

* Jack Steiner <steiner@sgi.com> wrote:

> Increase the maximum physical address size of x86_64 system to 
> 44-bits. This is in preparation for future chips that support larger 
> physical memory sizes.

thanks, applied.

	Ingo

--
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: [PATCH] - Increase max physical memory size of x86_64

[Andreas Herrmann]

On Fri, Mar 21, 2008 at 08:31:57AM -0500, Jack Steiner wrote:
> Increase the maximum physical address size of x86_64 system
> to 44-bits. This is in preparation for future chips that
> support larger physical memory sizes.

Shouldn't this be increased to 48?
AMD family 10h CPUs actually support 48 bits for the
physical address.


Regards,

Andreas


--
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: [PATCH] - Increase max physical memory size of x86_64

[Jack Steiner]

On Tue, Mar 25, 2008 at 05:41:54PM +0100, Andreas Herrmann wrote:
> On Fri, Mar 21, 2008 at 08:31:57AM -0500, Jack Steiner wrote:
> > Increase the maximum physical address size of x86_64 system
> > to 44-bits. This is in preparation for future chips that
> > support larger physical memory sizes.
> 
> Shouldn't this be increased to 48?
> AMD family 10h CPUs actually support 48 bits for the
> physical address.

You are probably correct but I don't work with AMD processors
and don't understand their requirements. If someone
wants to submit a patch to support larger phys memory sizes,
I certainly have no objections....


--- jack

--
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: [PATCH] - Increase max physical memory size of x86_64

[Chris Snook]

Jack Steiner wrote:
> On Tue, Mar 25, 2008 at 05:41:54PM +0100, Andreas Herrmann wrote:
>> On Fri, Mar 21, 2008 at 08:31:57AM -0500, Jack Steiner wrote:
>>> Increase the maximum physical address size of x86_64 system
>>> to 44-bits. This is in preparation for future chips that
>>> support larger physical memory sizes.
>> Shouldn't this be increased to 48?
>> AMD family 10h CPUs actually support 48 bits for the
>> physical address.
> 
> You are probably correct but I don't work with AMD processors
> and don't understand their requirements. If someone
> wants to submit a patch to support larger phys memory sizes,
> I certainly have no objections....

The only advantage 44 bits has over 48 bits is that it allows us to uniquely 
identify 4k physical pages with 32 bits, potentially allowing for tighter 
packing of certain structures.  Do we have any code that does this, and if so, 
is it a worthwhile optimization?

Personally, I think we should support the full capability of the hardware, but I 
don't have a 17 TB Opteron box to test with.

-- Chris

--
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: [PATCH] - Increase max physical memory size of x86_64

[Andreas Herrmann]

On Tue, Mar 25, 2008 at 05:02:46PM -0400, Chris Snook wrote:
> Jack Steiner wrote:
>> On Tue, Mar 25, 2008 at 05:41:54PM +0100, Andreas Herrmann wrote:
>>> On Fri, Mar 21, 2008 at 08:31:57AM -0500, Jack Steiner wrote:
>>>> Increase the maximum physical address size of x86_64 system
>>>> to 44-bits. This is in preparation for future chips that
>>>> support larger physical memory sizes.
>>> Shouldn't this be increased to 48?
>>> AMD family 10h CPUs actually support 48 bits for the
>>> physical address.
>> You are probably correct but I don't work with AMD processors
>> and don't understand their requirements. If someone
>> wants to submit a patch to support larger phys memory sizes,
>> I certainly have no objections....
>
> The only advantage 44 bits has over 48 bits is that it allows us to 
> uniquely identify 4k physical pages with 32 bits, potentially allowing for 
> tighter packing of certain structures.  Do we have any code that does this, 
> and if so, is it a worthwhile optimization?

I've checked where those defines are used. If I didn't miss something
MAX_PHYSADDR_BITS isn't used at all on x86 and MAX_PHYSMEM_BITS is
used (directly or indirectly) in several other macros.

But basically it's just section_to_node_table which would increase to 2
or 4 MB depending on MAX_NUMNODES.  Using 44 bits this table is just
128 kB resp. 256 kB in size.

> Personally, I think we should support the full capability of the hardware, 
> but I don't have a 17 TB Opteron box to test with.

I don't have one either.
By adjusting some NB-registers it might be possible to configure
physical addresses larger than 40 or 44 bits though. (Even if the
machine has not more than 1 or 16 TB.) I'll verify whether this is
really possible.

At the moment I think it's best to leave the define as is (44 or 40
bit) as there is currently no practical benefit from increasing it to
48 bit.


Regards,

Andreas


--
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: [PATCH] - Increase max physical memory size of x86_64

[Jack Steiner]

On Thu, Mar 27, 2008 at 06:30:27PM +0100, Andreas Herrmann wrote:
> On Tue, Mar 25, 2008 at 05:02:46PM -0400, Chris Snook wrote:
> > Jack Steiner wrote:
> >> On Tue, Mar 25, 2008 at 05:41:54PM +0100, Andreas Herrmann wrote:
> >>> On Fri, Mar 21, 2008 at 08:31:57AM -0500, Jack Steiner wrote:
> >>>> Increase the maximum physical address size of x86_64 system
> >>>> to 44-bits. This is in preparation for future chips that
> >>>> support larger physical memory sizes.
> >>> Shouldn't this be increased to 48?
> >>> AMD family 10h CPUs actually support 48 bits for the
> >>> physical address.
> >> You are probably correct but I don't work with AMD processors
> >> and don't understand their requirements. If someone
> >> wants to submit a patch to support larger phys memory sizes,
> >> I certainly have no objections....
> >
> > The only advantage 44 bits has over 48 bits is that it allows us to 
> > uniquely identify 4k physical pages with 32 bits, potentially allowing for 
> > tighter packing of certain structures.  Do we have any code that does this, 
> > and if so, is it a worthwhile optimization?
> 
> I've checked where those defines are used. If I didn't miss something
> MAX_PHYSADDR_BITS isn't used at all on x86 and MAX_PHYSMEM_BITS is
> used (directly or indirectly) in several other macros.
> 
> But basically it's just section_to_node_table which would increase to 2
> or 4 MB depending on MAX_NUMNODES.  Using 44 bits this table is just
> 128 kB resp. 256 kB in size.
> 
> > Personally, I think we should support the full capability of the hardware, 
> > but I don't have a 17 TB Opteron box to test with.
> 
> I don't have one either.
> By adjusting some NB-registers it might be possible to configure
> physical addresses larger than 40 or 44 bits though. (Even if the
> machine has not more than 1 or 16 TB.) I'll verify whether this is
> really possible.
> 
> At the moment I think it's best to leave the define as is (44 or 40
> bit) as there is currently no practical benefit from increasing it to
> 48 bit.

Sounds reasonable to me (44 bits). Let someone with access to
new hardware verify that changing to 48 actually works. 


--- jack

--
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>