In 2.5.5 (Changeset 1.359.2.3), the highmem.h and pgalloc.h includes were
reversed.
Previous kernels had linux/highmem.h including asm/pgalloc.h. Placing
the architecture cache handling functions worked well in asm/pgalloc.h,
allowing highmem.h to find them; highmem.h uses flush_dcache_page() and
friends in inline functions.
In this changeset, highmem.h no longer includes pgalloc.h, but instead
pgalloc.h includes highmem.h. This, unfortunately, tends to break things
in a major way since the cache functions are no longer available to
highmem.h.
Looking at x86, the (no-op) cache functions live in pgtable.h. However,
trying to put the ARM cache handling into pgtable.h doesn't work because
we need things like PG_arch_1 and struct page. pgtable.h is included by
linux/mm.h before linux/mm.h declares PG_arch_1 and struct page. So
obviously this can't work.
Has anyone encountered this, and has anyone found a magic working
combination?
For now, I'm going to reverse the include changes in this changeset so
things build again.
--
Russell King ([email protected]) The developer of ARM Linux
http://www.arm.linux.org.uk/personal/aboutme.html
On Thu, Feb 28, 2002 at 08:48:26PM +0000, Russell King wrote:
> For now, I'm going to reverse the include changes in this changeset so
> things build again.
That's what I did for alpha too.
r~
On February 28, 2002 09:48 pm, Russell King wrote:
> In 2.5.5 (Changeset 1.359.2.3), the highmem.h and pgalloc.h includes were
> reversed.
>
> Previous kernels had linux/highmem.h including asm/pgalloc.h. Placing
> the architecture cache handling functions worked well in asm/pgalloc.h,
> allowing highmem.h to find them; highmem.h uses flush_dcache_page() and
> friends in inline functions.
>
> In this changeset, highmem.h no longer includes pgalloc.h, but instead
> pgalloc.h includes highmem.h. This, unfortunately, tends to break things
> in a major way since the cache functions are no longer available to
> highmem.h.
>
> Looking at x86, the (no-op) cache functions live in pgtable.h. However,
> trying to put the ARM cache handling into pgtable.h doesn't work because
> we need things like PG_arch_1 and struct page. pgtable.h is included by
> linux/mm.h before linux/mm.h declares PG_arch_1 and struct page. So
> obviously this can't work.
>
> Has anyone encountered this, and has anyone found a magic working
> combination?
I've noticed it's a mess...
I ran into a different problem yesterday. It seems mmzone.h is included
before struct page is defined and therefore inlines in mmzone.h can't
do arithmetic on the size of struct page. This forces code in mmzone.h
to be macros instead of inlines.
--
Daniel