Non-ip packets such as ARP or LACP are not delivered if you use RTE flow rule
with empty pattern and rss action with specific combination of RSS hash types:
"ipv4-tcp ipv4-udp ipv4-other ipv6-tcp ipv6-udp ipv6-other esp".
The stack which used for RSS expansion was overflowed and trashed rss
expanstion data(buf->entry[MLX5_RSS_EXP_ELT_N]). Use bigger stack size instead.
And add relevant ASSERT for the future.
Signed-off-by: Maayan Kashani<mkash...@nvidia.com>
Acked-by: Ori Kam<or...@nvidia.com>
---
drivers/net/mlx5/mlx5_flow.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index cf83db7b60..41e298855b 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -374,7 +374,7 @@ mlx5_flow_expand_rss_skip_explicit(const struct
mlx5_flow_expand_node graph[],
return next;
}
-#define MLX5_RSS_EXP_ELT_N 16
+#define MLX5_RSS_EXP_ELT_N 32
/**
* Expand RSS flows into several possible flows according to the RSS hash
@@ -539,6 +539,7 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf,
size_t size,
if (lsize > size)
return -EINVAL;
n = elt * sizeof(*item);
+ MLX5_ASSERT((buf->entries) < MLX5_RSS_EXP_ELT_N);
buf->entry[buf->entries].priority =
stack_pos + 1 + missed;
buf->entry[buf->entries].pattern = addr;
--
2.25.1
