16K page size?

16K page size?

[Jun Sun]

Has anybody tried that before?

I wonder if it is just as simple as changing PAGE_SHIFT from 12 to 14 in
include/asm-mips/page.h.  

I remember I have seen many places in kernel and drivers that assume 4k
page size (perhaps minimum 4k page size in reality.)  What about glibc? 
Does it assume any page size?

Thanks.

Jun

Re: 16K page size?

[Ralf Baechle]

On Tue, Oct 17, 2000 at 11:18:28PM -0700, Jun Sun wrote:

> Has anybody tried that before?
> 
> I wonder if it is just as simple as changing PAGE_SHIFT from 12 to 14 in
> include/asm-mips/page.h.  
> 
> I remember I have seen many places in kernel and drivers that assume 4k
> page size (perhaps minimum 4k page size in reality.)  What about glibc? 
> Does it assume any page size?

Most applications probably use the getpagesize() function, so they should
be fine.  libc itself should also be clean.

In the kernel we don't handle this properly yet.  There are also some
optimizations which are possible for larger page sizes.  IA64 already
has a larger pagesize than Intel, so I hope they have already solve
most of the problems for us.

  Ralf

Re: 16K page size?

[Ralf Baechle]

On Wed, Oct 18, 2000 at 03:30:02AM +0200, Ralf Baechle wrote:

> Most applications probably use the getpagesize() function, so they should
> be fine.  libc itself should also be clean.
> 
> In the kernel we don't handle this properly yet.  There are also some
> optimizations which are possible for larger page sizes.  IA64 already
> has a larger pagesize than Intel, so I hope they have already solve
                             ^^^^^
                             i386
> most of the problems for us.

  Ralf

Re: 16K page size?

[Scott Venier]

shouldn't most things be safe since alpha uses 8k pages?

Scott

On Wed, 18 Oct 2000, Ralf Baechle wrote:

> On Wed, Oct 18, 2000 at 03:30:02AM +0200, Ralf Baechle wrote:
> 
> > Most applications probably use the getpagesize() function, so they should
> > be fine.  libc itself should also be clean.
> > 
> > In the kernel we don't handle this properly yet.  There are also some
> > optimizations which are possible for larger page sizes.  IA64 already
> > has a larger pagesize than Intel, so I hope they have already solve
>                              ^^^^^
>                              i386
> > most of the problems for us.
> 
>   Ralf
> 

Re: 16K page size?

[Ralf Baechle]

On Wed, Oct 18, 2000 at 10:24:43PM -0400, Scott Venier wrote:

> shouldn't most things be safe since alpha uses 8k pages?

I've seen places which had #ifdef __alpha__ instead of conditionals on
the pagesize ...

  Ralf