public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* x86 Page Sizes
@ 2002-06-26 22:02 Dan Sturtevant
  2002-06-26 22:11 ` Robert Love
  0 siblings, 1 reply; 7+ messages in thread
From: Dan Sturtevant @ 2002-06-26 22:02 UTC (permalink / raw)
  To: linux-kernel


I know the x86 linux kernel has 4K pages in userspace and 4M pages in 
kernel space.  These two sizes seem to be limitations of the intel 
architecture (I think).

Does anyone know a way to increase the userspace page size above 4K?
Are there any patches for a 4M userspace pagesize?

Thanks,
Dan


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-06-26 22:02 x86 Page Sizes Dan Sturtevant
@ 2002-06-26 22:11 ` Robert Love
  2002-06-27  6:35   ` Peter Svensson
  0 siblings, 1 reply; 7+ messages in thread
From: Robert Love @ 2002-06-26 22:11 UTC (permalink / raw)
  To: Dan Sturtevant; +Cc: linux-kernel

On Wed, 2002-06-26 at 18:02, Dan Sturtevant wrote:
> 
> I know the x86 linux kernel has 4K pages in userspace and 4M pages in 
> kernel space.  These two sizes seem to be limitations of the intel 
> architecture (I think).
> 
> Does anyone know a way to increase the userspace page size above 4K?
> Are there any patches for a 4M userspace pagesize?

Kernel has 4K pages in user and kernel space.  It is the same address
space and segment, just uses MMU protection.

x86 does 4K pages.

	Robert Love


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-06-26 22:11 ` Robert Love
@ 2002-06-27  6:35   ` Peter Svensson
  2002-06-27 14:26     ` Steven Cole
  2002-07-06  0:36     ` john slee
  0 siblings, 2 replies; 7+ messages in thread
From: Peter Svensson @ 2002-06-27  6:35 UTC (permalink / raw)
  To: Robert Love; +Cc: Dan Sturtevant, linux-kernel

On 26 Jun 2002, Robert Love wrote:

> Kernel has 4K pages in user and kernel space.  It is the same address
> space and segment, just uses MMU protection.
> 
> x86 does 4K pages.

The x86 cpus can use 4K or 4M pages in the hardware. The 4M pages are 
restricted to the kernel in Linux due to various problems. This has been 
discussed on this list a while ago. The thread was called "Have the 2.4 
kernel memory management problems on large machines been fixed?" the last 
time around.

4M pages are useful to minimize tlb misses which can be costly for some 
algorithms.

Peter
--
Peter Svensson      ! Pgp key available by finger, fingerprint:
<petersv@psv.nu>    ! 8A E9 20 98 C1 FF 43 E3  07 FD B9 0A 80 72 70 AF
------------------------------------------------------------------------
Remember, Luke, your source will be with you... always...



^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-06-27  6:35   ` Peter Svensson
@ 2002-06-27 14:26     ` Steven Cole
  2002-06-27 15:39       ` Robert Love
  2002-07-06  0:36     ` john slee
  1 sibling, 1 reply; 7+ messages in thread
From: Steven Cole @ 2002-06-27 14:26 UTC (permalink / raw)
  To: Peter Svensson; +Cc: Robert Love, Dan Sturtevant, linux-kernel

On Thu, 2002-06-27 at 00:35, Peter Svensson wrote:
> On 26 Jun 2002, Robert Love wrote:
> 
> > Kernel has 4K pages in user and kernel space.  It is the same address
> > space and segment, just uses MMU protection.
> > 
> > x86 does 4K pages.
> 
> The x86 cpus can use 4K or 4M pages in the hardware. The 4M pages are 
> restricted to the kernel in Linux due to various problems. This has been 
> discussed on this list a while ago. The thread was called "Have the 2.4 
> kernel memory management problems on large machines been fixed?" the last 
> time around.
> 
> 4M pages are useful to minimize tlb misses which can be costly for some 
> algorithms.
> 
> Peter

In addition to that thread, you might want to read the paper "Multiple
Page Size Support in the Linux Kernel", pages 573-593 in the Proceedings
of the Ottawa Linux Symposium which you can download from here:

http://www.linuxsymposium.org/2002/ 

If you're on a slow connection, be forewarned that it is a 631 page pdf,
but thankfully it's compressed.

Steven



^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-06-27 14:26     ` Steven Cole
@ 2002-06-27 15:39       ` Robert Love
  0 siblings, 0 replies; 7+ messages in thread
From: Robert Love @ 2002-06-27 15:39 UTC (permalink / raw)
  To: Steven Cole; +Cc: Peter Svensson, Dan Sturtevant, linux-kernel

On Thu, 2002-06-27 at 10:26, Steven Cole wrote:

> In addition to that thread, you might want to read the paper "Multiple
> Page Size Support in the Linux Kernel", pages 573-593 in the Proceedings
> of the Ottawa Linux Symposium which you can download from here:

Maybe I'll just go to the talk :)

	Robert Love



^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-06-27  6:35   ` Peter Svensson
  2002-06-27 14:26     ` Steven Cole
@ 2002-07-06  0:36     ` john slee
  2002-07-06  6:22       ` William Lee Irwin III
  1 sibling, 1 reply; 7+ messages in thread
From: john slee @ 2002-07-06  0:36 UTC (permalink / raw)
  To: Peter Svensson; +Cc: linux-kernel

On Thu, Jun 27, 2002 at 08:35:05AM +0200, Peter Svensson wrote:
> The x86 cpus can use 4K or 4M pages in the hardware. The 4M pages are 

DDJ ran an article quite a few years on this very topic.  i'm sure they
also mentioned that some processors (ppro/p2 onwards?) are capable of
2MiB pages

j.

-- 
toyota power: http://indigoid.net/

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x86 Page Sizes
  2002-07-06  0:36     ` john slee
@ 2002-07-06  6:22       ` William Lee Irwin III
  0 siblings, 0 replies; 7+ messages in thread
From: William Lee Irwin III @ 2002-07-06  6:22 UTC (permalink / raw)
  To: john slee; +Cc: Peter Svensson, linux-kernel

On Thu, Jun 27, 2002 at 08:35:05AM +0200, Peter Svensson wrote:
>> The x86 cpus can use 4K or 4M pages in the hardware. The 4M pages are 

On Sat, Jul 06, 2002 at 10:36:17AM +1000, john slee wrote:
> DDJ ran an article quite a few years on this very topic.  i'm sure they
> also mentioned that some processors (ppro/p2 onwards?) are capable of
> 2MiB pages

This has been rehashed too many times. There is a nice table in Intel's
processor manuals (the 3rd volumes on System Programming) describing
which combinations of options give which page sizes.


Cheers,
Bill

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2002-07-06  6:21 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-06-26 22:02 x86 Page Sizes Dan Sturtevant
2002-06-26 22:11 ` Robert Love
2002-06-27  6:35   ` Peter Svensson
2002-06-27 14:26     ` Steven Cole
2002-06-27 15:39       ` Robert Love
2002-07-06  0:36     ` john slee
2002-07-06  6:22       ` William Lee Irwin III

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox