Hi, On Tue, Oct 28, 2025 at 11:26:11AM +0200, Rongwei Liu wrote: > The memory footprint was increased due to the alignment.
s/alignment/overalignment/ For clarity that the relevant fields were overaligned. > > Signed-off-by: Rongwei Liu <[email protected]> > --- Since this is a bug fix, please add Fixes tag for the following commit: 820ca7361bb7 ("net/mlx5: fix flow aging race condition") Cc-ing to stable mailing list won't be needed, because the offending patch is only in 25.11-rc1 > drivers/net/mlx5/mlx5_hws_cnt.h | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/mlx5/mlx5_hws_cnt.h b/drivers/net/mlx5/mlx5_hws_cnt.h > index f5b7e8f643..d19a7bf37e 100644 > --- a/drivers/net/mlx5/mlx5_hws_cnt.h > +++ b/drivers/net/mlx5/mlx5_hws_cnt.h > @@ -43,7 +43,7 @@ struct mlx5_hws_cnt_dcs_mng { > }; > > union mlx5_hws_cnt_state { > - alignas(RTE_CACHE_LINE_SIZE) RTE_ATOMIC(uint32_t)data; > + RTE_ATOMIC(uint32_t)data; Please add a space between `RTE_ATOMIC(uint32_t)` and `data`. > struct { > uint32_t in_used:1; > /* Indicator whether this counter in used or in pool. */ > @@ -64,7 +64,7 @@ struct mlx5_hws_cnt { > struct flow_counter_stats reset; > union mlx5_hws_cnt_state cnt_state; > /* This struct is only meaningful when user own this counter. */ > - alignas(RTE_CACHE_LINE_SIZE) RTE_ATOMIC(uint32_t)query_gen_when_free; > + RTE_ATOMIC(uint32_t)query_gen_when_free; Same as above. > /* > * When PMD own this counter (user put back counter to PMD > * counter pool, i.e), this field recorded value of counter > -- > 2.27.0 > Best regards, Dariusz Sosnowski
