Re: [PATCH] Make CONFIG_MIGRATION available for s390

[Christoph Lameter <cl@linux-foundation.org>]

Gerald Schaefer wrote:
> On Mon, 2008-07-07 at 10:09 -0500, Christoph Lameter wrote:
>> This will extend the number of pages that are migratable and lead to strange
>> semantics in the NUMA case. There suddenly vma_is migratable will forbid hotplug
>> to migrate certain pages. 
>>
>> I think we need two functions:
>>
>> vma_migratable()	General migratability
>>
>> vma_policy_migratable()	Migratable under NUMA policies.
> 
> Nothing will change here for the NUMA case, this is all about making it
> compile w/o NUMA and with MIGRATION. What new strange semantics do you mean?
> BTW, the latest patch in this thread will not touch vma_migratable() anymore,
> I haven't read your mail before, sorry.

Ahh. Okay. However, we may still want a function that tells us if the pages in a vma
are migratable in general (independent of policies). The current vma_migratable function
tell you if the pages in a vma were placed according to a NUMA memory policy and should be migrated for NUMA locality optimizations.

>> That wont work since the migrate function takes a nodemask! The point of
>> the function is to move memory from node to node which is something that you
>> *cannot* do in a non NUMA configuration. So leave this chunk out.
> 
> Right, but I noticed that this function definition was needed to make it
> compile with MIGRATION and w/o NUMA, although it would never be called in
> non-NUMA config.

How does the compile break? It may be better to fix this where the function
is used.

> A better solution would probably be to put migrate_vmas(), the only caller
> of vm_ops->migrate(), inside '#ifdef CONFIG_NUMA', because it will only be
> called from NUMA-only mm/mempolicy.c. Does that sound reasonable?

That sounds right.
 
>> Hmmm... Okay. I tried to make MIGRATION as independent of CONFIG_NUMA as possible so hopefully this will work.
> 
> Umm, it doesn't compile with MIGRATION and w/o NUMA, which was the reason
> for this patch, because of the policy_zone reference in vma_migratable()
> and the missing vm_ops->migrate() function.

Right. I did not have a use case for !NUMA when I wrote the code. So you now need to
fix the minor bits that break.