empty_zero_page

empty_zero_page

[David S. Miller]

Why does mm/filemap_xip.c make an explicit reference to
"empty_zero_page"?  That's bogus, and ZERO_PAGE() is how
generic code should get at this thing.

In fact, what the mm/filemap_xip.c code wants is the page
struct, not the address of the page itself, because it
does a virt_to_page() on empty_zero_page in every such
reference.

This causes build failures for XIP support on sparc64.

When moving mm/filemap_xip.c over to ZERO_PAGE(), we will
need to determine the virtual address at which the ZERO_PAGE()
will be mapped.  This shouldn't be difficult to determine,
and it's incredibly important to get this right, wrt. page
coloring concerns, particularly on MIPS which does make use
of the 'vaddr' argument to ZERO_PAGE().

Re: empty_zero_page

[Gerald Schaefer]

On 7/5/05, David S. Miller <davem@davemloft.net> wrote:
> 
> Why does mm/filemap_xip.c make an explicit reference to
> "empty_zero_page"?  That's bogus, and ZERO_PAGE() is how
> generic code should get at this thing.
> 
> In fact, what the mm/filemap_xip.c code wants is the page
> struct, not the address of the page itself, because it
> does a virt_to_page() on empty_zero_page in every such
> reference.
> 
> This causes build failures for XIP support on sparc64.
> 
> When moving mm/filemap_xip.c over to ZERO_PAGE(), we will
> need to determine the virtual address at which the ZERO_PAGE()
> will be mapped.  This shouldn't be difficult to determine,
> and it's incredibly important to get this right, wrt. page
> coloring concerns, particularly on MIPS which does make use
> of the 'vaddr' argument to ZERO_PAGE().

Good point, seems like there is no reason to use empty_zero_page instead
of ZERO_PAGE. Carsten is out of the office this week, but we will get back
to this next week.