Re: [PATCH v3 1/2] mempool: add stack (lifo) mempool handler

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

On Mon, Jun 20, 2016 at 05:56:40PM +0000, Ananyev, Konstantin wrote:
> 
> 
> > -----Original Message-----
> > From: Jerin Jacob [mailto:jerin.jacob@caviumnetworks.com]
> > Sent: Monday, June 20, 2016 3:22 PM
> > To: Ananyev, Konstantin
> > Cc: Thomas Monjalon; dev@dpdk.org; Hunt, David; olivier.matz@6wind.com; viktorin@rehivetech.com; shreyansh.jain@nxp.com
> > Subject: Re: [dpdk-dev] [PATCH v3 1/2] mempool: add stack (lifo) mempool handler
> > 
> > On Mon, Jun 20, 2016 at 01:58:04PM +0000, Ananyev, Konstantin wrote:
> > >
> > >
> > > > -----Original Message-----
> > > > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Thomas Monjalon
> > > > Sent: Monday, June 20, 2016 2:54 PM
> > > > To: Jerin Jacob
> > > > Cc: dev@dpdk.org; Hunt, David; olivier.matz@6wind.com; viktorin@rehivetech.com; shreyansh.jain@nxp.com
> > > > Subject: Re: [dpdk-dev] [PATCH v3 1/2] mempool: add stack (lifo) mempool handler
> > > >
> > > > 2016-06-20 18:55, Jerin Jacob:
> > > > > On Mon, Jun 20, 2016 at 02:08:10PM +0100, David Hunt wrote:
> > > > > > This is a mempool handler that is useful for pipelining apps, where
> > > > > > the mempool cache doesn't really work - example, where we have one
> > > > > > core doing rx (and alloc), and another core doing Tx (and return). In
> > > > > > such a case, the mempool ring simply cycles through all the mbufs,
> > > > > > resulting in a LLC miss on every mbuf allocated when the number of
> > > > > > mbufs is large. A stack recycles buffers more effectively in this
> > > > > > case.
> > > > > >
> > > > > > Signed-off-by: David Hunt <david.hunt@intel.com>
> > > > > > ---
> > > > > >  lib/librte_mempool/Makefile            |   1 +
> > > > > >  lib/librte_mempool/rte_mempool_stack.c | 145 +++++++++++++++++++++++++++++++++
> > > > >
> > > > > How about moving new mempool handlers to drivers/mempool? (or similar).
> > > > > In future, adding HW specific handlers in lib/librte_mempool/ may be bad idea.
> > > >
> > > > You're probably right.
> > > > However we need to check and understand what a HW mempool handler will be.
> > > > I imagine the first of them will have to move handlers in drivers/
> > >
> > > Does it mean it we'll have to move mbuf into drivers too?
> > > Again other libs do use mempool too.
> > > Why not just lib/librte_mempool/arch/<arch_specific_dir_here>
> > > ?
> > 
> > I was proposing only to move only the new
> > handler(lib/librte_mempool/rte_mempool_stack.c). Not any library or any
> > other common code.
> > 
> > Just like DPDK crypto device, Even if it is software implementation its
> > better to move in driver/crypto instead of lib/librte_cryptodev
> > 
> > "lib/librte_mempool/arch/" is not correct place as it is platform specific
> > not architecture specific and HW mempool device may be PCIe or platform
> > device.
> 
> Ok, but why rte_mempool_stack.c has to be moved?

Just thought of having all the mempool handlers at one place.
We can't move all HW mempool handlers at lib/librte_mempool/

Jerin

> I can hardly imagine it is a 'platform sepcific'.
> From my understanding it is a generic code.
> Konstantin
> 
> 
> > 
> > > Konstantin
> > >
> > >
> > > > Jerin, are you volunteer?