Re: [dpdk-stable] [dpdk-dev] [PATCH] net: fix compilation with pedantic enabled

[Olivier Matz <olivier.matz@6wind.com>]

Hi Raslan,

On Tue, Jul 21, 2020 at 07:37:51AM +0000, Raslan Darawsheh wrote:
> Hi,
> > -----Original Message-----
> > From: Olivier Matz <olivier.matz@6wind.com>
> > Sent: Tuesday, July 21, 2020 10:09 AM
> > To: Ferruh Yigit <ferruh.yigit@intel.com>
> > Cc: Raslan Darawsheh <rasland@mellanox.com>; dev@dpdk.org;
> > stable@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH] net: fix compilation with pedantic enabled
> > 
> > Hi,
> > 
> > On Tue, Jul 21, 2020 at 01:05:57AM +0100, Ferruh Yigit wrote:
> > > On 7/16/2020 1:12 PM, Raslan Darawsheh wrote:
> > > > when trying to compile rte_mpls with pedantic enabled,
> > > > it will complain about bit field defintion.
> > > > error: type of bit-field 'bs' is a GCC extension [-Werror=pedantic]
> > > > error: type of bit-field 'tc' is a GCC extension [-Werror=pedantic]
> > > > error: type of bit-field 'tag_lsb' is a GCC extension [-Werror=pedantic]
> > > '
> > > I tried to reproduce by adding to '-pedantic' to 'rte_net.c' (which uses
> > > 'rte_mpls.h') but not able to get the warning. Is this happen with specific
> > > version of the compiler?
> 
> Yes It happens only with old compilers, maybe I should have mentioned that in the commit log (my mistake).
> gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-28)
> 
> > >
> > > >
> > > > This fixes the compilation error.
> > > >
> > > > Fixes: e480cf487a0d ("net: add MPLS header structure")
> > > > Cc: olivier.matz@6wind.com
> > > > Cc: stable@dpdk.org
> > > >
> > > > Signed-off-by: Raslan Darawsheh <rasland@mellanox.com>
> > > > ---
> > > >  lib/librte_net/rte_mpls.h | 12 ++++++------
> > > >  1 file changed, 6 insertions(+), 6 deletions(-)
> > > >
> > > > diff --git a/lib/librte_net/rte_mpls.h b/lib/librte_net/rte_mpls.h
> > > > index db91707..ecd1f64 100644
> > > > --- a/lib/librte_net/rte_mpls.h
> > > > +++ b/lib/librte_net/rte_mpls.h
> > > > @@ -24,13 +24,13 @@ extern "C" {
> > > >  struct rte_mpls_hdr {
> > > >  	uint16_t tag_msb;   /**< Label(msb). */
> > > >  #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
> > > > -	uint8_t tag_lsb:4;  /**< Label(lsb). */
> > > > -	uint8_t tc:3;       /**< Traffic class. */
> > > > -	uint8_t bs:1;       /**< Bottom of stack. */
> > > > +	uint32_t tag_lsb:4;  /**< Label(lsb). */
> > > > +	uint32_t tc:3;       /**< Traffic class. */
> > > > +	uint32_t bs:1;       /**< Bottom of stack. */
> > > >  #else
> > > > -	uint8_t bs:1;       /**< Bottom of stack. */
> > > > -	uint8_t tc:3;       /**< Traffic class. */
> > > > -	uint8_t tag_lsb:4;  /**< label(lsb) */
> > > > +	uint32_t bs:1;       /**< Bottom of stack. */
> > > > +	uint32_t tc:3;       /**< Traffic class. */
> > > > +	uint32_t tag_lsb:4;  /**< label(lsb) */
> > > >  #endif
> > > >  	uint8_t  ttl;       /**< Time to live. */
> > > >  } __rte_packed;
> > >
> > > The struct size keeps same after change, do you know if this behavior is
> > part of
> > > standard and guaranteed?
> > 
> > I have the same fear.
> To my understanding and please correct me if I'm wrong, the type of the bit fields shouldn't change the size of the structure,
> As long as the bit order is kept the same, and I made a small test for it and checked the size of the struct it gave 4 bytes (sizeof()) with both definitions.

You are probably right, however we saw some differences in the
behavior in specific conditions.
See https://patchwork.dpdk.org/patch/70458/ for instance.

> > 
> > Would it make sense to add __extension__ instead? We already do that
> > for gre, for instance.
> Yes I guess this can work as well,
> 
> Kindest regards
> Raslan Darawsheh