[dpdk-dev] What is the status of mempool manager?

[Jan Viktorin]

Hello Olivier...

On Tue, 17 May 2016 15:56:32 +0200
Olivier MATZ <olivier.matz at 6wind.com> wrote:

> Hi Jan,
> 
> On 05/17/2016 02:34 PM, Jan Viktorin wrote:
> > Hello,
> >
> > I was trying to find out the status of the mempool rework and found
> > this 36-pieces long patch set:
> >
> > http://dpdk.org/ml/archives/dev/2016-April/037464.html
> >
> > (I failed to apply it to 16.04, 2.1.0, current HEAD.)  
> 
> It should be integrated soon, as there was no comment during
> some time. I'll send a new version that applies on head
> + minor cosmetic changes.

Cool ;).

> 
> > and some bits around:
> >
> > http://dpdk.org/ml/archives/dev/2016-May/038390.html  
> 
> I need to review this one, it's on my todo list.
> 
> > http://dpdk.org/ml/archives/dev/2016-April/037509.html  
> 
> I think the code is ready to be integrated, but I was expecting
> some reactions on the v4, especially some precisions about the
> use cases and the needs. See in the link "Things that should
> still be discussed".
> 
> > http://dpdk.org/ml/archives/dev/2016-April/037457.html  
> 
> It is integrated since today ;)
> 
> > http://dpdk.org/ml/archives/dev/2016-April/036979.html  
> 
> Some comments have been done by Konstantin and myself, I
> think a new version will be submitted. It will be integrated
> in 16.07.

Thanks for the summary.

> 
> > ...but I am confused.
> >
> > I am trying to find out how to write a custom memory pool based on
> > the uio_dmem_genirq driver. It provides DMA memory via the UIO API
> > (maps).
> >
> > If I have a PMD running of top of this UIO, I need to implement a
> > custom allocator that gets memory from the dev->mem_resource (with
> > the mappings preloaded from the UIO by EAL).
> >
> > This is related to the SoC infra as given here:
> >
> > http://dpdk.org/ml/archives/dev/2016-May/038486.html
> >
> > Any idea how to start with this?  
> 
> I think your feedback would be very valuable.

OK, CC me.

> 
> The "external mempool manager" series allows to register new
> handler that replaces the internal ring storing the objects. It
> does not change the way the memory is allocated.
> 
> The big patchset introduce new way of allocating memory used by
> memory pool (no need to be virtually contiguous anymore). You
> could throw an eye on rte_mempool_populate*() functions that
> populate a mempool with memory for storing the objects.

Thanks for this hint. I've already noticed this function...

> 
> I'm not sure I'm getting your exact needs.
> Do you need to populate a ring-based mempool with specific memory,
> retrieved from UIO API?

Probably yes.

> Do you need to replace the default internal ring by something else?

I don't think so.

> Do you need to access the mempool from the hardware?

No.

Thanks
Jan

> 
> Regards,
> Olivier



-- 
   Jan Viktorin                  E-mail: Viktorin at RehiveTech.com
   System Architect              Web:    www.RehiveTech.com
   RehiveTech
   Brno, Czech Republic