Hey Marc,

On 02/14/2012 10:21 AM, Marc Dietrich wrote:

Hi Seth,

On Tuesday 14 February 2012 08:35:13 Seth Jennings wrote:

quoted

On 02/14/2012 05:39 AM, Marc Dietrich wrote:

quoted

Am 14.02.2012 12:22, schrieb Marc Dietrich:

quoted

Am 13.02.2012 15:47, schrieb Seth Jennings:

quoted

linux/vmalloc.h added to zsmalloc-main.c to resolve implicit
declaration errors.

X86 dependency added to zsmalloc and dependent drivers zcache and
zram.

This X86 only requirement is not ideal.  Working to find portable
functions for __flush_tlb_one and set_pte.

sorry, zcache/zram works fine one ARM here. Why not depend on
!POWERPC?

ups, sorry again, I tested an older version only. flush_tlb_one and
set_pte is only used with the new allocator as it seems.

It would seem that the cpu-local tlb page flushing function for ARM
is local_flush_tlb_kernel_page(), by look at the kmap_atomic() code
in arch/arm/mm/highmem.c.

So each arch is using a slightly different approach here.

A simple #define per-arch could make a it a common call.  I think
I'm going to code it up and see how it goes.

... and I'm happy to test it ;-)

So I have a branch ready for you to test if your are willing

git://github.com/spartacus06/linux.git -b portable

The branch is based on Greg's staging-next, since my
patches that remove the sparsemem dependency are in there.
The frontswap_v11 patches are in that branch as well.

I did a simple make test with my ARM cross compiler and it
builds cleanly.  Now I just need to know if it _works_.

I'm not sure if you have been experimenting with frontswap
since it's not in mainline yet, but you can enable/disable
it in:

Kernel Features -> Enable frontswap to cache swap pages if tmem is present

Only the frontswap paths use zsmalloc right now. We are looking
to use it for the cleancache paths too soon (I'm working on
that now).

I'd appreciate any feedback you can provide!

Thanks,
Seth