Re: [PATCH net-next v2] net: mlx5: Allow police as the last action of a TC rule

From: Mohsin Bashir

Date: Fri Nov 20 2026 - 17:00:07 EST




On 11/20/26 1:14 PM, Mark Bloch wrote:


On 19/11/2026 6:00, Mohsin Bashir wrote:
From: Mohsin Bashir <hmohsin@xxxxxxxx>

In switchdev mode, mlx5 offloads police only when other actions follow
it. TC needs no extra action for a police-only rule to work. The mlx5
driver makes every police rule forward to the post meter table, but for
a rate police it adds that forward only after verify_attr_actions() has
checked the rule, so the check rejects it. The ok result of police is
also handled as pipe. In TC, ok accepts the packet and ends the action
list, whereas mlx5 goes on with the actions after police.

Support police as the only action of a rule. Set the forward while
parsing police, as is already done for police with mtu, and forward
an ok result to the end of the TC chains, as the driver does when it
offloads the gact ok action. This also changes existing offloaded rules
where police is followed by other actions. Before, an ok result ran the
actions after police. Now it accepts the packet, as TC does. The
upcoming selftest series also relies on such rules for TC offload
validation.

When police is the last action, a pipe result also ends the action
list, so forward it to the end of the TC chains too. Otherwise it is
sent back to the post action table, and with two police actions the
packet loops through the second police. A jump counts from the start
of the action list, so its target may be an earlier action or police
itself. Reject jump when police is the last action.

The patch was tested on a ConnectX-7 in switchdev mode with the
commands below. It was verified, using the hardware drop counter, that
the rules are offloaded and limit the matching traffic to the
configured rate, that the second police of the two police rule only
sees the packets conforming to the first, and that jump on the last
police is rejected.

tc filter add dev eth0 ingress pref 100 protocol ipv6 flower skip_sw \
ip_proto udp dst_port 5201 \
action police rate 100mbit burst 1m conform-exceed drop/ok

tc filter add dev eth0 ingress pref 100 protocol ipv6 flower skip_sw \
ip_proto udp dst_port 5201 \
action police rate 200mbit burst 1m conform-exceed drop/pipe \
action police rate 100mbit burst 1m conform-exceed drop/pipe

tc filter add dev eth0 ingress pref 100 protocol ipv6 flower skip_sw \
ip_proto udp dst_port 5201 \
action police rate 100mbit burst 1m conform-exceed drop / jump 1

Cc: Carolina Jubran <cjubran@xxxxxxxxxx>
Signed-off-by: Mohsin Bashir <hmohsin@xxxxxxxx>
---
Changelog:
v2:
- Treat pipe on the last police as ok and forward it to the end of the
TC chains.
- Reject jump when police is the last action, since the jump target is
counted from the start of the action list.
- Update commit message

v1: https://lore.kernel.org/netdev/20261117031337.4126403-1-mohsin.bashr@xxxxxxxxx

I think we can keep this patch small and tighten the JUMP checks:

for each police branch using JUMP:
reject unless police is the first action
reject unless the target is a later, existing action
reject if skipped actions include anything other than DROP/OK
reject if the countdown would count two separate actions as one
(e.g. repeated uses of a shared action)

That keeps the simple forward jumps working and leaves the new OK/PIPE
support intact. It rejects some valid configurations, but that seems
better than offloading them with different behavior.

Broader JUMP support would need more work on the action parser to make
sure hardware behaves the same as software. I'd handle that separately.

Unless there's a concrete use case, I'd keep backward jumps unsupported.
We don't want to offload steering loops, intentionally or by mistake.

I'm not sure which JUMP cases you need. Would this be enough for your
use case?

Mark

Thanks Mark, I believe this should works for us (at-least for now). We mainly need police as the only action and pipe or ok on the conform side, so for now we don't need jump beyond what's supported today. I'll add these checks in v3,


---
.../mellanox/mlx5/core/en/tc/act/police.c | 11 +++++++-
.../net/ethernet/mellanox/mlx5/core/en_tc.c | 26 ++++++++++++-------
2 files changed, 27 insertions(+), 10 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c
index 1bd1c94fb977..faa0822b1136 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c
@@ -49,6 +49,14 @@ tc_act_can_offload_police(struct mlx5e_tc_act_parse_state *parse_state,
if (err)
return false;
+ if (flow_action_is_last_entry(parse_state->flow_action, act) &&
+ (act->police.exceed.act_id == FLOW_ACTION_JUMP ||
+ act->police.notexceed.act_id == FLOW_ACTION_JUMP)) {
+ NL_SET_ERR_MSG_MOD(parse_state->extack,
+ "Jump on last police action is not supported");
+ return false;
+ }
+
return !!mlx5e_get_flow_meters(parse_state->flow->priv->mdev);
}
@@ -94,13 +102,14 @@ tc_act_parse_police(struct mlx5e_tc_act_parse_state *parse_state,
MLX5_FLOW_STEERING_CAP_MATCH_RANGES))
return -EOPNOTSUPP;
- attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
attr->flags |= MLX5_ATTR_FLAG_MTU;
} else {
attr->action |= MLX5_FLOW_CONTEXT_ACTION_EXECUTE_ASO;
attr->exe_aso_type = MLX5_EXE_ASO_FLOW_METER;
}
+ attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
+
return 0;
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index 1d35c97aea0b..7e7111e6d058 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -3874,7 +3874,7 @@ static int
alloc_branch_attr(struct mlx5e_tc_flow *flow,
struct mlx5e_tc_act_branch_ctrl *cond,
struct mlx5_flow_attr **cond_attr,
- u32 *jump_count,
+ u32 *jump_count, bool last,
struct netlink_ext_ack *extack)
{
struct mlx5_flow_attr *attr;
@@ -3891,11 +3891,17 @@ alloc_branch_attr(struct mlx5e_tc_flow *flow,
case FLOW_ACTION_DROP:
attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
break;
- case FLOW_ACTION_ACCEPT:
case FLOW_ACTION_PIPE:
- err = set_branch_dest_ft(flow->priv, attr);
- if (err)
- goto out_err;
+ if (!last) {
+ err = set_branch_dest_ft(flow->priv, attr);
+ if (err)
+ goto out_err;
+ break;
+ }
+ fallthrough;
+ case FLOW_ACTION_ACCEPT:
+ attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
+ attr->flags |= MLX5_ATTR_FLAG_ACCEPT;
break;
case FLOW_ACTION_JUMP:
if (*jump_count) {
@@ -3967,7 +3973,7 @@ dec_jump_count(struct flow_action_entry *act, struct mlx5e_tc_act *tc_act,
static int
parse_branch_ctrl(struct flow_action_entry *act, struct mlx5e_tc_act *tc_act,
struct mlx5e_tc_flow *flow, struct mlx5_flow_attr *attr,
- struct mlx5e_tc_jump_state *jump_state,
+ struct mlx5e_tc_jump_state *jump_state, bool last,
struct netlink_ext_ack *extack)
{
struct mlx5e_tc_act_branch_ctrl cond_true, cond_false;
@@ -3980,7 +3986,7 @@ parse_branch_ctrl(struct flow_action_entry *act, struct mlx5e_tc_act *tc_act,
tc_act->get_branch_ctrl(act, &cond_true, &cond_false);
err = alloc_branch_attr(flow, &cond_true,
- &attr->branch_true, &jump_count, extack);
+ &attr->branch_true, &jump_count, last, extack);
if (err)
goto out_err;
@@ -3988,7 +3994,7 @@ parse_branch_ctrl(struct flow_action_entry *act, struct mlx5e_tc_act *tc_act,
jump_state->jumping_attr = attr->branch_true;
err = alloc_branch_attr(flow, &cond_false,
- &attr->branch_false, &jump_count, extack);
+ &attr->branch_false, &jump_count, last, extack);
if (err)
goto err_branch_false;
@@ -4052,7 +4058,9 @@ parse_tc_actions(struct mlx5e_tc_act_parse_state *parse_state,
dec_jump_count(act, tc_act, attr, priv, &jump_state);
- err = parse_branch_ctrl(act, tc_act, flow, attr, &jump_state, extack);
+ err = parse_branch_ctrl(act, tc_act, flow, attr, &jump_state,
+ flow_action_is_last_entry(flow_action, act),
+ extack);
if (err)
goto out_free_post_acts;