Re: [dpdk-dev] [PATCH v2] mbuf: display more fields in dump

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

On Fri, Dec 27, 2019 at 09:05:07AM -0800, Stephen Hemminger wrote:
> On Fri, 27 Dec 2019 10:10:18 +0100
> Olivier Matz <olivier.matz@6wind.com> wrote:
> 
> > Hi,
> > 
> > On Thu, Dec 26, 2019 at 08:58:51AM -0800, Stephen Hemminger wrote:
> > > On Thu, 26 Dec 2019 17:15:39 +0100
> > > Olivier Matz <olivier.matz@6wind.com> wrote:
> > >   
> > > > Hi,
> > > > 
> > > > On Thu, Nov 21, 2019 at 10:30:55AM -0800, Stephen Hemminger wrote:  
> > > > > The rte_pktmbuf_dump should display offset, refcount, and vlan
> > > > > info since these are often useful during debugging.
> > > > > 
> > > > > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> > > > > ---
> > > > > v2 - remove casts, change in_port to port
> > > > >      the refcount and offset are property of per-segment
> > > > > 
> > > > >  lib/librte_mbuf/rte_mbuf.c | 17 ++++++++++-------
> > > > >  1 file changed, 10 insertions(+), 7 deletions(-)
> > > > > 
> > > > > diff --git a/lib/librte_mbuf/rte_mbuf.c b/lib/librte_mbuf/rte_mbuf.c
> > > > > index 35df1c4c38a5..4894d46628e3 100644
> > > > > --- a/lib/librte_mbuf/rte_mbuf.c
> > > > > +++ b/lib/librte_mbuf/rte_mbuf.c
> > > > > @@ -473,18 +473,21 @@ rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len)
> > > > >  
> > > > >  	__rte_mbuf_sanity_check(m, 1);
> > > > >  
> > > > > -	fprintf(f, "dump mbuf at %p, iova=%"PRIx64", buf_len=%u\n",
> > > > > -	       m, (uint64_t)m->buf_iova, (unsigned)m->buf_len);
> > > > > -	fprintf(f, "  pkt_len=%"PRIu32", ol_flags=%"PRIx64", nb_segs=%u, "
> > > > > -	       "in_port=%u\n", m->pkt_len, m->ol_flags,
> > > > > -	       (unsigned)m->nb_segs, (unsigned)m->port);
> > > > > +	fprintf(f, "dump mbuf at %p, iova=%#"PRIx64", buf_len=%u\n",
> > > > > +		m, m->buf_iova, m->buf_len);
> > > > > +	fprintf(f,
> > > > > +		"  pkt_len=%u, ol_flags=%#"PRIx64", nb_segs=%u, port=%u, vlan_tci=%#x\n",
> > > > > +		m->pkt_len, m->ol_flags, m->nb_segs, m->port, m->vlan_tci);
> > > > > +
> > > > >  	nb_segs = m->nb_segs;
> > > > >  
> > > > >  	while (m && nb_segs != 0) {
> > > > >  		__rte_mbuf_sanity_check(m, 0);
> > > > >  
> > > > > -		fprintf(f, "  segment at %p, data=%p, data_len=%u\n",
> > > > > -			m, rte_pktmbuf_mtod(m, void *), (unsigned)m->data_len);
> > > > > +		fprintf(f, "  segment at %p, data=%p, len=%u, off=%u, refcnt=%u\n",
> > > > > +			m, rte_pktmbuf_mtod(m, void *),
> > > > > +			m->data_len, m->data_off, rte_mbuf_refcnt_read(m));
> > > > > +
> > > > >  		len = dump_len;
> > > > >  		if (len > m->data_len)
> > > > >  			len = m->data_len;
> > > > > -- 
> > > > > 2.20.1
> > > > >     
> > > > 
> > > > Thanks for this enhancement.
> > > > 
> > > > One comment however: I don't see why vlan_tci would be important than
> > > > packet type or rss tag. It is also a bit confusing to display a field
> > > > which can have a random value (if the vlan flag is not set in ol_flags).
> > > > 
> > > > I'd prefer to dump vlan_tci only if the flag is present. Later we could
> > > > add more fields with the same logic. What do you think?  
> > > 
> > > Because this is a debugging hook and easier to always display the value.
> > > Thats all.  
> > 
> > This is precisely because it is a debug hook that I find dangerous to
> > display a wrong value. As the mbuf are recycled, a mbuf that does not
> > have a vlan_tci can display its previous value, which may look valid.
> > 
> > What about simply doing this?
> > 
> > 	fprintf(f,
> > 		"  pkt_len=%u, ol_flags=%#"PRIx64", nb_segs=%u, port=%u",
> > 		m->pkt_len, m->ol_flags, m->nb_segs, m->port);
> > 	if (m->ol_flags & PKT_RX_VLAN)
> > 		fprintf(f, ", vlan_tci=%#x", m->vlan_tci);
> > 	fprintf(f, "\n");
> > 
> > If you want I can submit an updated version of your patch
> 
> The right flags are more complex, you need to deal with TX vlan.
> And PKT_RX_VLAN means "this is a vlan packet" it doesn't mean
> vlan tag is stripped into TCI.  That is PKT_RX_VLAN_STRIPPED

PKT_RX_VLAN means that m->vlan_tci is valid, and this is what we
want here.