[dpdk-dev] [PATCH] eal: promote some service core functions to stable

[Neil Horman]

On Sat, Jun 22, 2019 at 07:51:10PM +0200, David Marchand wrote:
> On Sat, Jun 22, 2019 at 6:17 PM Neil Horman <nhorman at tuxdriver.com> wrote:
> 
> > On Fri, Jun 21, 2019 at 09:58:41PM +0200, David Marchand wrote:
> > > On Fri, Jun 21, 2019 at 7:41 PM Neil Horman <nhorman at tuxdriver.com>
> > wrote:
> > >
> > > > On Fri, Jun 21, 2019 at 06:47:31PM +0200, David Marchand wrote:
> > > > > On Fri, Jun 21, 2019 at 6:28 PM Neil Horman <nhorman at tuxdriver.com>
> > > > wrote:
> > > > >
> > > > > > On Fri, Jun 21, 2019 at 02:45:45PM +0200, David Marchand wrote:
> > > > > > > Ok, did a new pass on the tree.. found quite some sites where we
> > have
> > > > > > > issues (and other discrepancies... I started a new patchset).
> > > > > > > Looked at gcc documentation [1], and to me the safer approach
> > would
> > > > be to
> > > > > > > enforce that __rte_experimental is the first thing of a symbol
> > > > > > declaration.
> > > > > > >
> > > > > > > Comments?
> > > > > > >
> > > > > > Yes, thats the only way it works, in fact I'm suprised gcc didn't
> > > > throw an
> > > > > > error
> > > > > > about expecting an asm statement if you put it anywhere else
> > > > > >
> > > > >
> > > > > - I tried this, but then I hit issues with inlines.
> > > > > Like for example:
> > > > >
> > > > > static inline char * __rte_experimental
> > > > > rte_mbuf_buf_addr(struct rte_mbuf *mb, struct rte_mempool *mp)
> > > > > {
> > > > >   return (char *)mb + sizeof(*mb) + rte_pktmbuf_priv_size(mp);
> > > > > }
> > > > >
> > > > > I did not find a way to move the __rte_experimental tag without
> > getting
> > > > > warnings.
> > > > Right, thats the way its supposed to work on gcc/icc/clang.  function
> > > > attributes
> > > > must be declared between the return type and the function name,
> > anything
> > > > else
> > > > will generate compiler warnings/errors.  Because __rte_experimental
> > > > expands to a
> > > > __attribute__(...), you have to place it there.
> > > >
> > > > > If I try to compile some sources which includes rte_mbuf.h but
> > without
> > > > > -DALLOW_EXPERIMENTAL_API, then gcc errors at including the header,
> > > > > complaining that rte_mbuf_buf_addr() is deprecated, even if this
> > inline
> > > > is
> > > > > not called.
> > > > >
> > > > Thats...odd.  I wonder if thats an artifact of the function being
> > marked as
> > > > inline.  The compiler is supposed to insert the warning for any
> > remaining
> > > > calls
> > > > after dead code eliminitaion.  If the function is inline, I wonder if
> > the
> > > > compiler conservatively inserts the warning because it got expanded
> > into
> > > > another
> > > > function, when it can't tell if it will be entirely elimintated.  Can
> > you
> > > > provide a code sample that demonstrates this?
> > > >
> > > >
> > > rte_mbuf_buf_addr() is called in rte_mbuf_data_addr_default(), both of
> > them
> > > are unused by the includers of rte_mbuf.h.
> > >
> > >
> > > Reproduced it like this:
> > >
> > > [dmarchan at dmarchan ~]$ cat deprecated.c
> > > __attribute__((deprecated)) static inline void *plap(void)
> > > {
> > > return 0;
> > > }
> > >
> > > __attribute__((deprecated)) static inline void *plep(void)
> > > {
> > > plap();
> > > return 0;
> > > }
> > >
> > > int main(int argc, char *argv[])
> > > {
> > > return 0;
> > > }
> > > [dmarchan at dmarchan ~]$ gcc -o deprecated -Wall deprecated.c
> > > deprecated.c: In function ‘plep’:
> > > deprecated.c:8:2: warning: ‘plap’ is deprecated (declared at
> > > deprecated.c:1) [-Wdeprecated-declarations]
> > >   plap();
> > >   ^
> > >
> > Hmm, yes, that seems buggy to me.  I wonder if you are seeing this bug in
> > action:
> >
> > https://gcc.gnu.org/bugzilla//show_bug.cgi?id=80680
> 
> 
> It has the same flavor yes.
> Currently using gcc version 4.8.5 20150623 (Red Hat 4.8.5-36) (GCC)
> 
> 
> 
> >
> > Seem like the behavior fits.  It would be interesting to know if clang and
> > icc
> > suffer from the same issue
> >
> 
> Just tried, clang is fine.
> clang version 3.4.2 (tags/RELEASE_34/dot2-final)
> 
> 
> Actually, I went and protected this call to rte_mbuf_buf_addr().
> And with just this, it builds fine.
> I think I am going to take this approach, just a little comment :-).
> 
Thats probably the best workaround for this at the moment, I agree.  I'll add a
comment to the gcc bug and see if we can't get some movement on it

thanks
Neil

> 
> -- 
> David Marchand