Xen devel - Sep 2009 - Synchronous scrubbing for alloc_domheap_pages()

I''m running into an issue where alloc_domheap_pages() fails, even
though there is free memory, because the memory hasn''t been scrubbed
yet.  The context of this is when ballooning down one VM in order to
start a second VM: the balloon driver on the first VM has reached its
target, but creation of the second VM fails because the freed memory
hasn''t been scrubbed yet.

In the situation where there is no memory on the free list, but memory
on the "to-be-srubbed" list, is there any reason why we don''t
do
syncronous scrubbing?

The alternative would be to teach the control tools to look at
unscrubbed vs free memory, and wait until there was enough to start
the VM; but that seems a bit more hackish.

I chatted with Tim Deegan about this, and the only issue he could see
would be making sure that hypercalls which called
alloc_domheap_pages() yielded reasonably.

Thoughts?
 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Cancel that... the real problem was that set_pod_target() was trying
to allocate superpages, but not falling back to 4k pages if that
failed.

 -George

On Wed, Sep 23, 2009 at 3:08 PM, George Dunlap
<George.Dunlap@eu.citrix.com> wrote:
> I''m running into an issue where alloc_domheap_pages() fails, even
> though there is free memory, because the memory hasn''t been
scrubbed
> yet.  The context of this is when ballooning down one VM in order to
> start a second VM: the balloon driver on the first VM has reached its
> target, but creation of the second VM fails because the freed memory
> hasn''t been scrubbed yet.
>
> In the situation where there is no memory on the free list, but memory
> on the "to-be-srubbed" list, is there any reason why we
don''t do
> syncronous scrubbing?
>
> The alternative would be to teach the control tools to look at
> unscrubbed vs free memory, and wait until there was enough to start
> the VM; but that seems a bit more hackish.
>
> I chatted with Tim Deegan about this, and the only issue he could see
> would be making sure that hypercalls which called
> alloc_domheap_pages() yielded reasonably.
>
> Thoughts?
>  -George
>
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

And in fact page scrubbing *is* entirely synchronous in xen-unstable now
(but not in any stable branch, as yet).

 -- Keir

On 23/09/2009 15:40, "George Dunlap"
<George.Dunlap@eu.citrix.com> wrote:
> Cancel that... the real problem was that set_pod_target() was trying
> to allocate superpages, but not falling back to 4k pages if that
> failed.
> 
>  -George
> 
> On Wed, Sep 23, 2009 at 3:08 PM, George Dunlap
> <George.Dunlap@eu.citrix.com> wrote:
>> I''m running into an issue where alloc_domheap_pages() fails,
even
>> though there is free memory, because the memory hasn''t been
scrubbed
>> yet.  The context of this is when ballooning down one VM in order to
>> start a second VM: the balloon driver on the first VM has reached its
>> target, but creation of the second VM fails because the freed memory
>> hasn''t been scrubbed yet.
>> 
>> In the situation where there is no memory on the free list, but memory
>> on the "to-be-srubbed" list, is there any reason why we
don''t do
>> syncronous scrubbing?
>> 
>> The alternative would be to teach the control tools to look at
>> unscrubbed vs free memory, and wait until there was enough to start
>> the VM; but that seems a bit more hackish.
>> 
>> I chatted with Tim Deegan about this, and the only issue he could see
>> would be making sure that hypercalls which called
>> alloc_domheap_pages() yielded reasonably.
>> 
>> Thoughts?
>>  -George
>> 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel