Re: [PATCH] allow vma merging with mlock et. al.

[Darren Hart <dvhltc@us.ibm.com>]

Chris Wright wrote:
> * Chris Wright (chrisw@osdl.org) wrote:
> 
>>* Darren Hart (dvhltc@us.ibm.com) wrote:
>>
>>>The were a couple long standing (since at least 2.4.21) superfluous 
>>>variables and two unnecessary assignments in do_mlock().  The intent of 
>>>the resulting code is also more obvious.
>>>
>>>Tested on a 4 way x86 box running a simple mlock test program.  No 
>>>problems detected.
>>
>>Did you test with multiple page ranges, and locking subsets of vmas?
>>Seems that splitting could cause a problem since you now sample vm_end
>>before and after fixup, where the vma could be changed in the middle.
> 
> 
> Actually I think it winds up being fine since we don't do merging with
> mlock.  But why not?  Patch below remedies that.

We don't merge, but we do split if necessary, so the temp variables are 
still needed.  As I understand it, the reason we don't merge is because 
it is expected that a task will lock and unlock the same memory range 
more than once and we don't want to waste our time merging and splitting 
the VMAs.

Thanks,

--Darren