On 04/12/2014 15:42, Ananyev, Konstantin wrote: > > >> -----Original Message----- >> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Bruce Richardson >> Sent: Thursday, December 04, 2014 2:40 PM >> To: Jean-Mickael Guerin >> Cc: dev at dpdk.org >> Subject: Re: [dpdk-dev] [PATCH 1/2] ixgbe: fix setup of mbuf initializer >> template >> >> On Thu, Dec 04, 2014 at 03:26:20PM +0100, Jean-Mickael Guerin wrote: >>> Add a compiler barrier to make sure all fields covered by >>> the marker rearm_data are assigned before the read. >>> >>> Signed-off-by: Jean-Mickael Guerin <jean-mickael.guerin at 6wind.com> >>> Acked-by: David Marchand <david.marchand at 6wind.com> >>> Fixes: 0ff3324da2 ("ixgbe: rework vector pmd following mbuf changes") >> >> Acked-by: Bruce Richardson <bruce.richardson at intel.com> >> >>> --- >>> lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c | 3 +++ >>> 1 file changed, 3 insertions(+) >>> >>> diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c >>> b/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c >>> index 579bc46..c1b5a78 100644 >>> --- a/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c >>> +++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx_vec.c >>> @@ -739,6 +739,9 @@ ixgbe_rxq_vec_setup(struct igb_rx_queue *rxq) >>> mb_def.buf_len = rxq->mb_pool->elt_size - sizeof(struct rte_mbuf); >>> mb_def.port = rxq->port_id; >>> rte_mbuf_refcnt_set(&mb_def, 1); >>> + >>> + /* prevent compiler reordering: rearm_data covers previous fields */ >>> + rte_compiler_barrier(); >>> rxq->mbuf_initializer = *((uint64_t *)&mb_def.rearm_data); >>> return 0; >>> } >>> -- > > Hmm, can someone explain to me why do we need a compiler barrier here? > Konstantin
rearm_data is a separate field and as well an array of length zero, overlapping on purpose the fields data_off buf_len, port, refcnt. It might depend on compiler, but I could see a wrong value of 0UL for mbuf_initializer without the barrier (gcc 4.4.6). > >>> 2.1.3 >>>
