Re: Active Memory Defragmentation: Our implementation & problems

["Martin J. Bligh" <mbligh@xxxxxxxxxxx>]

>> There are a couple of special cases that might be feasible without making
>> an ungodly mess. PTE pages spring to mind (particularly as they can be
>> in highmem too). They should be reasonably easy to move (assuming we can
>> use rmap to track them back to the process they belong to to lock them ...
>> hmmm ....)
> 
> We don't do any pte page reclaim at any time other than process exit and
> there are plenty of pte pages we can just plain free anyway.  Anthing
> that's completely mapping page cache, for instance.
> 
> In the replacement case, taking mm->page_table_lock, doing the copy, and
> replacing the pointer from the pmd should be all that it takes.  But, I
> wonder if we could miss any sets of the pte dirty bit this way...

As long as we make sure the process doesn't run during the move, I don't
see why it'd be a problem. But I am less than convinced that rmap will
lead us back from the PTE page to the mm, at least w/o modification.

M.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>