Re: [dpdk-dev] [PATCH v2 2/5] eal: add accessor functions for lcore_config

[Stephen Hemminger <stephen@networkplumber.org>]

On Tue, 16 Apr 2019 17:03:47 +0000
Jerin Jacob Kollanukkaran <jerinj@marvell.com> wrote:

> > -----Original Message-----
> > From: dev <dev-bounces@dpdk.org> On Behalf Of Stephen Hemminger
> > Sent: Wednesday, April 10, 2019 10:46 PM
> > To: dev@dpdk.org
> > Cc: Stephen Hemminger <stephen@networkplumber.org>
> > Subject: [dpdk-dev] [PATCH v2 2/5] eal: add accessor functions for lcore_config
> > 
> > The fields of the internal EAL core configuration are currently laid bare as part of
> > the API. This is not good practice and limits fixing issues with layout and sizes.
> > 
> > Make new accessor functions for the fields used by current drivers and
> > examples. Mark return code functions as experimental since this value might
> > change in the future and probably shouldn't have been used by non EAL code
> > anyway.
> > 
> > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> > Reviewed-by: David Marchand <david.marchand@redhat.com>
> >  Shared Library Versions
> >  -----------------------
> > diff --git a/lib/librte_eal/common/eal_common_lcore.c
> > b/lib/librte_eal/common/eal_common_lcore.c
> > index 1cbac42286ba..6cf4d7abb0bd 100644
> > --- a/lib/librte_eal/common/eal_common_lcore.c
> > +++ b/lib/librte_eal/common/eal_common_lcore.c
> > @@ -16,6 +16,45 @@
> >  #include "eal_private.h"
> >  #include "eal_thread.h"
> > 
> > +int rte_lcore_index(int lcore_id)
> > +{
> > +	if (unlikely(lcore_id >= RTE_MAX_LCORE))
> > +		return -1;
> > +
> > +	if (lcore_id < 0)
> > +		lcore_id = (int)rte_lcore_id();
> > +
> > +	return lcore_config[lcore_id].core_index;
> > +}  
> 
> If I understand it correctly, We are planning to do this scheme only for slow path functions. Right?
> Is rte_lcore_* functions comes in slow path category? I thought a few of them can be used
> In fast path too.
> I am bit concerned about a low end arm cores where function invocation overhead significant vs inline.
> ODP has taken a route of introducing a compile time config to choose inline vs separate function to address
> the performance regression . I am not sure what would be the correct way to handle this.

The lcore config itself is not accessed anywhere in fastpath of any of the examples.
It is usually is part of the setup process. The rte_lcore_index is only used
by lthread in a log message.

The main fastpath is rte_lcore_id() which is already done by per-thread variable
and is unchanged by these patches.

I don't have facilities to do any deep dive performance testing on ARM.