[dpdk-dev] [PATCH v3 05/11] baseband/acc100: add LDPC processing functions

[Ananyev, Konstantin]

> > > > > > +
> > > > > > +static inline char *
> > > > > > +mbuf_append(struct rte_mbuf *m_head, struct rte_mbuf *m,
> > > > > > +uint16_t
> > > > > > +len) {
> > > > > > +	if (unlikely(len > rte_pktmbuf_tailroom(m)))
> > > > > > +		return NULL;
> > > > > > +
> > > > > > +	char *tail = (char *)m->buf_addr + m->data_off + m-
> > >data_len;
> > > > > > +	m->data_len = (uint16_t)(m->data_len + len);
> > > > > > +	m_head->pkt_len  = (m_head->pkt_len + len);
> > > > > > +	return tail;
> > > > > > +}
> > > > >
> > > > > Is it reasonable to direct add data_len of rte_mbuf?
> > > > >
> > > >
> > > > Do you suggest to add directly without checking there is enough room
> > > > in the mbuf? We cannot rely on the application providing mbuf with
> > > > enough tailroom.
> > >
> > > What I mentioned is this changes about mbuf should move to librte_mbuf.
> > > And it's better to align Olivier Matz.
> >
> > There is already rte_pktmbuf_append() inside rte_mbuf.h.
> > Wouldn't it suit?
> >
> 
> Hi Ananyev, Rosen,
> I agree that this can be confusing at first look and notably compared to packet processing.
> Note first that this same existing syntaxwhich  is already used in all bbdev PMDs when manipulating outbound mbufs in the context of base
> band signal processing (not really a packet as for NIC or other devices).
> Nothing new in that very PMD as this follows existing logic already in DPDK bbdev PMDs.
> 
> This function basically differs from the typical rte_pktmbuf_append() as this is not appending data in the last mbuf but is used to potentially
> update sequentially data for any mbufs in the middle from preallocated data hence it takes 2 arguments for both the head and the current
> mbuf segment in the list.

Ok, thanks for explanation.

> There may be a more elegant way to do this down the line notably once there is a proposal to handle gracefully large mbufs (another
> usecase we have to handle in a slightly custom way). But I believe that is orthogonal to that very PMD serie which keeps on reling on using
> existing logic.
> 
> 
> 
> 
> > >
> > > > In case you ask about the 2 mbufs, this is because this function is
> > > > used to also support segmented memory made of multiple mbufs segments.
> > > > Note that this function is also used in other existing bbdev PMDs.
> > > > In case you believe there is a better way to do this, we can
> > > > certainly discuss and change these in several PMDs through another serie.
> > > >
> > > > Thanks for all the reviews and useful comments.
> > > > Nic