On Mon, Mar 27, 2023 at 07:07:12AM +0000, Joyce Kong wrote:
> The following warning is observed with GCC12 compilation
> with release 20.11:
>
> In function ‘__rte_ring_enqueue_elems_64’,
> inlined from ‘__rte_ring_enqueue_elems’ at
> ../lib/librte_ring/rte_ring_elem.h:225:3,
> inlined from ‘__rte_ring_do_enqueue_elem’ at
> ../lib/librte_ring/rte_ring_elem.h:424:2,
> inlined from ‘rte_ring_mp_enqueue_burst_elem’ at
> ../lib/librte_ring/rte_ring_elem.h:884:9,
> inlined from ‘rte_ring_enqueue_burst_elem’ at
> ../lib/librte_ring/rte_ring_elem.h:946:10,
> inlined from ‘rte_ring_enqueue_burst’ at
> ../lib/librte_ring/rte_ring.h:721:9,
> inlined from ‘pdump_copy’ at
> ../lib/librte_pdump/rte_pdump.c:94:13:
> ../lib/librte_ring/rte_ring_elem.h:162:40: warning: ‘*dup_bufs.36_42
> + _89’ may be used uninitialized [-Wmaybe-uninitialized]
> 162 | ring[idx] = obj[i];
> | ~~~^~~
> ../lib/librte_ring/rte_ring_elem.h:163:44: warning: ‘*dup_bufs.36_42
> + _98’ may be used uninitialized [-Wmaybe-uninitialized]
> 163 | ring[idx + 1] = obj[i + 1];
> | ~~~^~~~~~~
> ../lib/librte_ring/rte_ring_elem.h:164:44: warning: ‘*dup_bufs.36_42
> + _107’ may be used uninitialized [-Wmaybe-uninitialized]
> 164 | ring[idx + 2] = obj[i + 2];
> | ~~~^~~~~~~
> ../lib/librte_ring/rte_ring_elem.h:165:44: warning: ‘*dup_bufs.36_42
> + _116’ may be used uninitialized [-Wmaybe-uninitialized]
> 165 | ring[idx + 3] = obj[i + 3];
> | ~~~^~~~~~~
> ../lib/librte_ring/rte_ring_elem.h:169:42: warning: ‘*dup_bufs.36_42
> + _129’ may be used uninitialized [-Wmaybe-uninitialized]
> 169 | ring[idx++] = obj[i++]; /* fallthrough */
> | ~~~^~~~~
> ../lib/librte_ring/rte_ring_elem.h:171:42: warning: ‘*dup_bufs.36_42
> + _139’ may be used uninitialized [-Wmaybe-uninitialized]
> 171 | ring[idx++] = obj[i++]; /* fallthrough */
> | ~~~^~~~~
> ../lib/librte_ring/rte_ring_elem.h:173:42: warning: ‘*dup_bufs.36_42
> + _149’ may be used uninitialized [-Wmaybe-uninitialized]
> 173 | ring[idx++] = obj[i++];
>
> Actually, this is an alias warning as -O3 enables strict alias.
> This patch fixes it by replacing 'dup_bufs' with '&dup_bufs[0]'
> as the compiler represents them differently.
>
> Fixes: 278f945402c5 ("pdump: add new library for packet capture")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Joyce Kong <joyce.k...@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.w...@arm.com>
> ---
Acked-by: Tyler Retzlaff <roret...@linux.microsoft.com>
> lib/pdump/rte_pdump.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/lib/pdump/rte_pdump.c b/lib/pdump/rte_pdump.c
> index 9bc4bab4f2..53cca1034d 100644
> --- a/lib/pdump/rte_pdump.c
> +++ b/lib/pdump/rte_pdump.c
> @@ -134,7 +134,7 @@ pdump_copy(uint16_t port_id, uint16_t queue,
>
> __atomic_fetch_add(&stats->accepted, d_pkts, __ATOMIC_RELAXED);
>
> - ring_enq = rte_ring_enqueue_burst(ring, (void *)dup_bufs, d_pkts, NULL);
> + ring_enq = rte_ring_enqueue_burst(ring, (void *)&dup_bufs[0], d_pkts,
> NULL);
nit: i would drop the cast to void *, it shouldn't be needed?
