Re: qemu-kvm defunct due to THP [was: mmotm 2011-01-06-15-41 uploaded]

[Jiri Slaby <jirislaby@gmail.com>]

On 01/10/2011 04:01 PM, Andrea Arcangeli wrote:
> On Mon, Jan 10, 2011 at 03:37:57PM +0100, Jiri Slaby wrote:
>> On 01/07/2011 12:41 AM, akpm@linux-foundation.org wrote:
>>> The mm-of-the-moment snapshot 2011-01-06-15-41 has been uploaded to
>>
>> Hi, something of the following breaks qemu-kvm:
> 
> Thanks for the report. It's already fixed and I posted this a few days
> ago to linux-mm.
> 
> I had to rewrite the KVM THP support when merging THP in -mm, because
> the kvm code in -mm has async page faults and doing so I eliminated
> one gfn_to_page lookup for each kvm secondary mmu page fault. But
> first new attempt wasn't entirely successful ;), the below incremental
> fix should work. Please test it and let me know if any trouble is
> left.
> 
> Also note again on linux-mm I posted two more patches, I recommend to
> apply the other two as well. The second adds KSM THP support, the
> third cleanup some code but I like to have it tested.
> 
> Thanks a lot,
> Andrea
> 
> ====
> Subject: thp: fix for KVM THP support
> 
> From: Andrea Arcangeli <aarcange@redhat.com>
> 
> There were several bugs: dirty_bitmap ignored (migration shutoff largepages),
> has_wrprotect_page(directory_level) ignored, refcount taken on tail page and
> refcount released on pfn head page post-adjustment (now it's being transferred
> during the adjustment, that's where KSM over THP tripped inside
> split_huge_page, the rest I found it by code review).
> 
> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
> ---
>  arch/x86/kvm/mmu.c         |   97 ++++++++++++++++++++++++++++++++-------------
>  arch/x86/kvm/paging_tmpl.h |   10 +++-
>  2 files changed, 79 insertions(+), 28 deletions(-)

Yup, this works for me. If you point me to the other 2, I will test them
too...

thanks,
-- 
js