lkml.org 
[lkml]   [2008]   [Jan]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [kvm-devel] [PATCH] export notifier #1
Christoph Lameter wrote:
> Ahhh. Good to hear. But we will still end in a situation where only
> the remote ptes point to the page. Maybe the remote instance will dirty
> the page at that point?
>
>

When the spte is dropped, its dirty bit is transferred to the page.
>
>> sharing code, and for you missing a single notifier means memory
>> corruption because you don't bump the page count to represent the
>> external reference).
>>
>
> The approach with the export notifier is page based not based on the
> mm_struct. We only need a single page count for a page that is exported to
> a number of remote instances of linux. The page count is dropped when all
> the remote instances have unmapped the page.
>

That won't work for kvm. If we have a hundred virtual machines, that
means 99 no-op notifications.

Also, our rmap key for finding the spte is keyed on (mm, va). I imagine
most RDMA cards are similar.
>
>
>>> @@ -966,6 +973,9 @@ int try_to_unmap(struct page *page, int
>>>
>>> BUG_ON(!PageLocked(page));
>>>
>>> + if (unlikely(PageExported(page)))
>>> + export_notifier(invalidate_page, page);
>>> +
>>>
>> Passing the page here will complicate things especially for shared
>> pages across different VM that are already working in KVM. For non
>>
>
> How?
>
>
>> shared pages we could cache the userland mapping address in
>> page->private but it's a kludge only working for non-shared
>> pages. Walking twice the anon_vma lists when only a single walk is
>>
>
> There is only the need to walk twice for pages that are marked Exported.
> And the double walk is only necessary if the exporter does not have its
> own rmap. The cross partition thing that we are doing has such an rmap and
> its a matter of walking the exporters rmap to clear out the external
> references and then we walk the local rmaps. All once.
>

The problem is that external mmus need a reverse mapping structure to
locate their ptes. We can't expand struct page so we need to base it on
mm + va.

>
>> Besides the pinned pages ram leak by having the zero locking window
>> above I'm curious how you are going to take care of the finegrined
>> aging that I'm doing with the accessed bit set by hardware in the spte
>>
>
> I think I explained that above. Remote users effectively are forbidden to
> establish new references to the page by the clearing of the exported bit.
>
>

Can they wait on that bit?


--
error compiling committee.c: too many arguments to function



\
 
 \ /
  Last update: 2008-01-23 11:31    [from the cache]
©2003-2011 Jasper Spaans