[dpdk-dev] [PATCH] eal: added new `rte_lcore_is_service_lcore` API.

Pavan Nikhilesh Bhagavatula pbhagavatula at caviumnetworks.com
Fri Sep 15 16:59:34 CEST 2017


On Fri, Sep 15, 2017 at 02:44:57PM +0000, Van Haaren, Harry wrote:
>
>
> > -----Original Message-----
> > From: Pavan Nikhilesh Bhagavatula [mailto:pbhagavatula at caviumnetworks.com]
> > Sent: Friday, September 15, 2017 3:42 PM
> > To: Van Haaren, Harry <harry.van.haaren at intel.com>
> > Cc: thomas at monjalon.net; dev at dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH] eal: added new `rte_lcore_is_service_lcore`
> > API.
> >
> > On Fri, Sep 15, 2017 at 01:57:42PM +0000, Van Haaren, Harry wrote:
> > > > From: Thomas Monjalon [mailto:thomas at monjalon.net]
> > > > Sent: Friday, September 15, 2017 2:53 PM
> > > > To: Pavan Nikhilesh Bhagavatula <pbhagavatula at caviumnetworks.com>; Van
> > Haaren,
> > > > Harry <harry.van.haaren at intel.com>
> > > > Cc: dev at dpdk.org
> > > > Subject: Re: [dpdk-dev] [PATCH] eal: added new `rte_lcore_is_service_lcore`
> > > > API.
> > > >
> > > > 28/08/2017 17:09, Pavan Nikhilesh Bhagavatula:
> > > > > On Mon, Aug 28, 2017 at 01:49:37PM +0000, Van Haaren, Harry wrote:
> > > > > > From: Pavan Nikhilesh Bhagavatula
> > [mailto:pbhagavatula at caviumnetworks.com]
> > > > > > > On Mon, Aug 28, 2017 at 10:59:51AM +0000, Van Haaren, Harry wrote:
> > > > > > > > From: Pavan Nikhilesh [mailto:pbhagavatula at caviumnetworks.com]
> > > > > > > > > --- a/lib/librte_eal/common/include/rte_lcore.h
> > > > > > > > > +++ b/lib/librte_eal/common/include/rte_lcore.h
> > > > > > > > > @@ -180,6 +180,24 @@ rte_lcore_is_enabled(unsigned lcore_id)
> > > > > > > > >  }
> > > > > > > > >
> > > > > > > > >  /**
> > > > > > > > > + * Test if an lcore is service lcore.
> > > > > > > > > + *
> > > > > > > > > + * @param lcore_id
> > > > > > > > > + *   The identifier of the lcore, which MUST be between 0 and
> > > > > > > > > + *   RTE_MAX_LCORE-1.
> > > > > > > > > + * @return
> > > > > > > > > + *   True if the given lcore is service; false otherwise.
> > > > > > > > > + */
> > > > > > > > > +static inline int
> > > > > > > > > +rte_lcore_is_service_lcore(unsigned lcore_id)
> > > > > > > > > +{
> > > > > > > > > +	struct rte_config *cfg = rte_eal_get_configuration();
> > > > > > > > > +	if (lcore_id >= RTE_MAX_LCORE)
> > > > > > > > > +		return 0;
> > > > > > > > > +	return cfg->lcore_role[lcore_id] == ROLE_SERVICE;
> > > > > > > > > +}
> > > > > > > >
> > > > > > > > No header file and Static inline - so this is only to be used
> > > > internally in the service
> > > > > > > cores library?
> > > > > > > > If so, the function should actually be used, instead of only added
> > but
> > > > not used in the
> > > > > > > library itself.
> > > > > > > >
> > > > > > >
> > > > > > > The enum rte_lcore_role_t has ROLE_SERVICE which tells that a
> > particular
> > > > lcore is
> > > > > > > a service lcore as well as an EAL thread some libraries such as
> > rte_timer
> > > > allow
> > > > > > > specific operations only over EAL threads.
> > > > > >
> > > > > > Understood that role of cores is important, and that rte_timer might
> > > > require this information.
> > > > > >
> > > > > >
> > > > > > > The rte_timer lib uses the rte_is_lcore_enabled() call to check if a
> > > > lcore is an
> > > > > > > EAL thread, Which checks if the lcore role is  ROLE_RTE. But it
> > should
> > > > also
> > > > > > > allow timers to be registered on a service core as processing those
> > > > timers can
> > > > > > > be done on them.
> > > > > >
> > > > > > No problem from me here either - although it's the Timers library
> > > > maintainer that should check this.
> > > > > >
> > > > > >
> > > > > > > This new function allows such libraries to check if the role is
> > > > > > > ROLE_SERVICE and allow those operations.
> > > > > >
> > > > > > If the timers library requires information about service-cores, it
> > should
> > > > use a public API to retrieve that information. Having "internal" functions
> > > > between libraries is not nice.
> > > > > >
> > > > > > I think a better design would be to add this function as a public
> > function,
> > > > (add it to the .map files etc) and then call the public function from the
> > > > timers library.
> > > > > >
> > > > > > Does that sound like a good solution? -Harry
> > > > > >
> > > > >
> > > > > The file rte_lcore.h is in librte_eal/common/include I couldn't find a
> > .map
> > > > > file for eal/common and also other functions that are present in
> > rte_lcore.h
> > > > > aren't mapped in eal/linuxapp or eal/bsdapp.
> > > > > I think it is fine as the functions are static inline.
> > > >
> > > > We must avoid adding more inline functions without a good justification.
> > > > The inline functions are tolerated for performance reasons only.
> > > >
> > > > We could also choose to add this function to rte_service.h ?
> > >
> > > Yes that is an option, and OK with me.
> > >
> > > @Pavan what do you think of adding it to service.h, implement in .c and add
> > to .map?
> > >
> >
> > The ROLE_SERVICE/ROLE_RTE defines the role of a lcore so it made sense to put
> > it in rte_lcore.h as lcore properties are accessed mostly through this header.
> > I'm fine with adding it to service.h as suggested by Harry.
> >
> > -Pavan
>
> *as suggested by Thomas ;)
>
> Initially I thought it made more sense in lcore.h too, however the application
> should only require knowing if core X is a service core if it cares about
> services / service-cores, hence I'm fine with rte_service.h too.
>
> -Harry
>
Agreed, will spin up a v2.

Thanks,
Pavan
>
>
>


More information about the dev mailing list