On Wed, May 18, 2022 at 12:16:48PM +0200, David Marchand wrote:
> GCC 12 raises the following warning:
>
> In function ‘__rte_ring_enqueue_elems_64’,
> inlined from ‘__rte_ring_enqueue_elems’ at
> ../lib/ring/rte_ring_elem_pvt.h:130:3,
> inlined from ‘__rte_ring_do_hts_enqueue_elem’ at
> ../lib/ring/rte_ring_hts_elem_pvt.h:196:3,
> inlined from ‘rte_ring_mp_hts_enqueue_burst_elem’ at
> ../lib/ring/rte_ring_hts.h:110:9,
> inlined from ‘rte_ring_enqueue_burst_elem’ at
> ../lib/ring/rte_ring_elem.h:577:10,
> inlined from ‘rte_ring_enqueue_burst’ at
> ../lib/ring/rte_ring.h:738:9,
> inlined from ‘process_op_bit’ at
> ../drivers/crypto/ipsec_mb/pmd_snow3g.c:425:16,
> inlined from ‘snow3g_pmd_dequeue_burst’ at
> ../drivers/crypto/ipsec_mb/pmd_snow3g.c:484:20:
> ../lib/ring/rte_ring_elem_pvt.h:68:44: error: array subscript 1 is
> outside array bounds of ‘struct rte_crypto_op[0]’
> [-Werror=array-bounds]
> 68 | ring[idx + 1] = obj[i + 1];
> | ~~~^~~~~~~
> ../drivers/crypto/ipsec_mb/pmd_snow3g.c: In function
> ‘snow3g_pmd_dequeue_burst’:
> ../drivers/crypto/ipsec_mb/pmd_snow3g.c:434:1: note:
> at offset 8 into object ‘op’ of size 8
> 434 | snow3g_pmd_dequeue_burst(void *queue_pair,
> | ^~~~~~~~~~~~~~~~~~~~~~~~
>
> Validate that one (exactly) op has been processed or return early.
>
> Fixes: b537abdbee74 ("crypto/snow3g: support bit-level operations")
> Cc: sta...@dpdk.org
>
> Signed-off-by: David Marchand <david.march...@redhat.com>
> ---
> drivers/crypto/ipsec_mb/pmd_snow3g.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/crypto/ipsec_mb/pmd_snow3g.c
> b/drivers/crypto/ipsec_mb/pmd_snow3g.c
> index ebc9a0b562..9a85f46721 100644
> --- a/drivers/crypto/ipsec_mb/pmd_snow3g.c
> +++ b/drivers/crypto/ipsec_mb/pmd_snow3g.c
> @@ -422,12 +422,13 @@ process_op_bit(struct rte_crypto_op *op, struct
> snow3g_session *session,
> op->sym->session = NULL;
> }
>
> - enqueued_op = rte_ring_enqueue_burst(qp->ingress_queue,
> - (void **)&op, processed_op, NULL);
> + if (unlikely(processed_op != 1))
> + return 0;
> + enqueued_op = rte_ring_enqueue(qp->ingress_queue, op);
As a fix for the compiler warning this looks ok, but question for
maintainer would be - should this check for processed_op != 1 not go
earlier in the function, immediately after the switch statement?
Fan, Pablo, can you please comment?
/Bruce
