[dpdk-dev] [PATCH 2/2] eal: rename key opaque pointer in TLS API

[Morten Brørup]

> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Tal Shnaiderman
> Sent: Tuesday, March 2, 2021 2:13 PM
> 
> > Subject: RE: [dpdk-dev] [PATCH 2/2] eal: rename key opaque pointer in
> TLS
> > API
> >
> > External email: Use caution opening links or attachments
> >
> >
> > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Tal
> Shnaiderman
> > > Sent: Tuesday, March 2, 2021 12:25 PM
> > >
> > > rename the key opaque pointer from rte_tls_key to
> rte_thread_tls_key
> > > to avoid confusion with transport layer security.
> > >
> >
> > [...]
> >
> > > diff --git a/lib/librte_eal/include/rte_thread.h
> > > b/lib/librte_eal/include/rte_thread.h
> > > index 39737d1829..90bcb02554 100644
> > > --- a/lib/librte_eal/include/rte_thread.h
> > > +++ b/lib/librte_eal/include/rte_thread.h
> > > @@ -23,7 +23,7 @@ extern "C" {
> > >  /**
> > >   * TLS key type, an opaque pointer.
> > >   */
> > > -typedef struct eal_tls_key *rte_tls_key;
> > > +typedef struct eal_tls_key *rte_thread_tls_key;
> > >
> > >  /**
> > >   * Set core affinity of the current thread.
> > > @@ -63,7 +63,8 @@ void rte_thread_get_affinity(rte_cpuset_t
> *cpusetp);
> > >   */
> > >
> > >  __rte_experimental
> > > -int rte_thread_tls_key_create(rte_tls_key *key, void
> > > (*destructor)(void *));
> > > +int rte_thread_tls_key_create(rte_thread_tls_key *key,
> > > +                     void (*destructor)(void *));
> > >
> >
> > I agree with your argument for TLS confusion.
> >
> > How about rte_thread_key, instead of rte_thread_tls_key. Having both
> > thread and tls seems redundant.
> >
> 
> Thanks for the input, make sense, I'll change the name to your
> suggestion.
> 
> >
> > Here are some more thoughts... It is meant as a provocation only, not
> a real
> > suggestion:
> >
> > The DPDK API often uses the term "lcore" as the abstraction for
> threads, e.g.
> > rte_per_lcore.h refers to thread local storage using "per_lcore",
> while it is in
> > fact "per thread". Why use another terminology in the API for thread
> keys,
> > instead of sticking with the "lcore" naming tradition, e.g. struct
> > rte_lcore_key?
> >
> 
> You're right, but then there are some functions in eal_common_thread.c
> which uses the 'thread' terminology, maybe it's a good idea to rework
> it all to a single accepted term.
> 

Agreed. Especially after the introduction of service cores and non-EAL thread support, using "lcore" as a terminology for "thread" seems to be a legacy convention with decreasing relevance.

However, cleaning up that convention would totally break the ABI, and take quite an effort. So I guess it will remain wishful thinking only. :-)

But we can follow your lead and do it right with new ABIs.