Re: [dpdk-dev] [PATCH] net/liquidio: do not touch mbuf initialized fields

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

Hi Ferruh,

On Fri, 9 Jun 2017 12:24:56 +0100, Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> On 6/8/2017 12:22 PM, Shijith Thotton wrote:
> > Avoid re-initializing of mbuf fields which are set while in pool.
> > Replaced lio_recv_buffer_alloc with rte_pktmbuf_alloc.
> > 
> > See commit 8f094a9ac5d7 ("mbuf: set mbuf fields while in pool").
> > 
> > Signed-off-by: Shijith Thotton <shijith.thotton@caviumnetworks.com>  
> 
> <...>
> 
> > @@ -489,9 +466,6 @@
> >  			droq->refill_count++;
> >  
> >  			if (likely(nicbuf != NULL)) {
> > -				nicbuf->data_off = RTE_PKTMBUF_HEADROOM;  
> 
> Hi Shijith, Olivier,
> 
> nb_segs and next are initialized properly when mbuf allocated, but is
> "data_off = RTE_PKTMBUF_HEADROOM" guaranteed?
> 
> > -				nicbuf->nb_segs = 1;
> > -				nicbuf->next = NULL;
> >  				/* We don't have a way to pass flags yet */
> >  				nicbuf->ol_flags = 0;
> >  				if (rh->r_dh.has_hash) {  
> 
> <...>

It's not guaranteed when using a raw allocation, i.e. rte_mbuf_raw_alloc()
or rte_mempool_get(). The PMD usually use these functions in Rx path to
fully initialize the mbuf fields on their own.

In this driver, it looks it uses rte_pktmbuf_alloc(), which calls
rte_pktmbuf_reset(), fully initializing the mbuf struct. So data_off is
properly set.


As a side note, I wonder if lio_recv_buffer_alloc() replaced by
rte_pktmbuf_alloc() everywhere in the code.


Olivier