Skip to contentSkip to conversation

ice: fix DFLT Rx rule handling for promisc and switchdev

PATCH Superseded Petr Oros 2 participants 5 messages Jun 18

Subsystems (3)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
author
#1 lore raw

PATCH 1/2 ice: skip per-VLAN promisc rules when default VSI Rx rule is set

In reply to Petr Oros

When an ice port is part of a vlan-filtering bridge with a wide VLAN trunk and the netdev is in IFF_PROMISC (typical for bond slaves attached to a bridge), the driver installs per-VLAN ICE_SW_LKUP_PROMISC_VLAN entries (recipe 9) in addition to the broad ICE_SW_LKUP_DFLT VSI Rx rule (recipe 5). Each per-VLAN rule consumes one Flow Lookup Unit (FLU) entry from a fixed hardware pool of "up to 32K FLU entries" per device, documented in the E810 datasheet (613875-009 section 7.8.10, Table 7-18, page 1015).

With three active PFs sharing one switch context and a bridge trunk of vid 2-4094, the configuration would require roughly

3 PFs * 4093 VLANs * 3 rules per VLAN per PF ~= 36,800 rules

which exceeds the 32K FLU budget. Firmware then responds to further Add Switch Rules requests with AQ retval 0x10 (LIBIE_AQ_RC_ENOSPC) and the user-visible failure surfaces as

ice 0000:5c:00.1: Failed to set VSI 14 as the default forwarding
VSI, error -5
ice 0000:5c:00.1 ens1f1: Error -5 setting default VSI 14 Rx rule

After a switch context has been driven into overrun, subsequent retries can come back as AQ retval 0x2 (LIBIE_AQ_RC_ENOENT), which has misled triage attempts toward a perceived recipe binding defect rather than a capacity issue.

When the DFLT VSI Rx rule is in place it catches every packet on the lport regardless of VLAN tag, so the per-VLAN PROMISC_VLAN expansion is redundant. The recipe 4 VLAN prune entries are still installed per VLAN and continue to track the allowed VID set, but the IFF_PROMISC sync path disables their enforcement on the VSI via vlan_ops->dis_rx_filtering() before ice_set_promisc() runs. ena_rx_filtering() is restored when IFF_PROMISC is cleared.

Skip the per-VLAN expansion at the two call sites that drive it:
ice_set_promisc() falls through to ice_fltr_set_vsi_promisc() and ice_vlan_rx_add_vid() omits the per-VLAN ICE_MCAST_VLAN_PROMISC_BITS add. Plain IFF_ALLMULTI without an installed DFLT VSI rule is unchanged and still installs per-VLAN multicast promisc rules.

Both checks use ice_is_vsi_dflt_vsi() which inspects the recipe filter list for an installed DFLT rule on this VSI, not netdev->flags & IFF_PROMISC. The HW-state predicate avoids two regression vectors that a user-intent predicate would introduce:

1. ice_lag_is_switchdev_running() short-circuits ice_set_dflt_vsi() to return 0 without installing the DFLT rule for a PF in switchdev LAG mode. An IFF_PROMISC-only check would also suppress the per-VLAN fallback, leaving the PF with no rule.

2. When ice_set_dflt_vsi() returns a non-EEXIST error (FLU exhausted, switch context divergence), the driver clears IFF_PROMISC from vsi->current_netdev_flags but the netdev's own flags retain IFF_PROMISC. The user-intent predicate would still suppress the per-VLAN fallback even though DFLT failed to install.

The predicate is install-time only. The IFF_PROMISC off path closes the lifecycle gap in ice_vsi_exit_dflt_promisc(): for an IFF_ALLMULTI VSI with VLANs it reinstates the per-VID rules before clearing the default rule, so multicast coverage never lapses. If that AQ call fails the default rule is left in place, ice_vsi_exit_dflt_promisc() returns the error, and the sync_fltr pass bails with vsi->current_netdev_flags |= IFF_PROMISC; the current/netdev flag mismatch re-fires the IFF_PROMISC off path on the next sync. Clearing the default rule first would instead expose a window where neither the default rule nor the per-VID rules carry multicast.

If ice_clear_dflt_vsi() fails after the per-VID rules were reinstated they are deliberately not rolled back. Clearing the default rule is a removal that frees an FLU entry rather than allocating one, so it cannot fail for lack of space; a failure is a transient AdminQ error. The per-VID rules are the steady state for an IFF_ALLMULTI VLAN VSI, so the only redundant entry left behind is the single un-removed default rule, not the per-VID set. The retry re-enters this path, ice_fltr_set_vlan_vsi_promisc() returns -EEXIST for the rules that already exist so nothing is reallocated, and the default rule is removed on the next attempt. Rolling the per-VID rules back here would instead churn thousands of removes and re-adds on every retry.

After the default rule is gone the vid=0 PROMISC rule that paired with it is redundant and is dropped, but only to reclaim a filter entry, so a failure there is logged and does not abort the transition.

ice_set_vsi_promisc() and ice_clear_vsi_promisc() dispatch the recipe based on whether ICE_PROMISC_VLAN_RX/TX bits are present in the mask: with the bits set, recipe ICE_SW_LKUP_PROMISC_VLAN is used; otherwise ICE_SW_LKUP_PROMISC. The else branch in ice_set_promisc() installs the vid=0 rule in ICE_SW_LKUP_PROMISC. Because ice_clear_promisc() with VLANs present adds the VLAN bits and would search ICE_SW_LKUP_PROMISC_VLAN, the recipe mismatch would leave the vid=0 ICE_SW_LKUP_PROMISC rule orphaned when VLANs are configured. This is a single stale rule, not a per-cycle leak:
re-adding it on the next promisc on returns -EEXIST rather than allocating a new entry. The set-time recipe is not recorded, so ice_clear_promisc() clears both recipes; clearing a rule that is not present succeeds, both clears run unconditionally, and the first error is returned.

The two VLAN-0 recipe transition blocks in ice_vlan_rx_add_vid() and ice_vlan_rx_kill_vid() that promote / demote the vid=0 rule between ICE_SW_LKUP_PROMISC and ICE_SW_LKUP_PROMISC_VLAN are likewise guarded by !ice_is_vsi_dflt_vsi(). With DFLT in place the vid=0 rule already covers every VID and a recipe swap would only install a redundant rule.

Lab reproduction on an E810-C with the same firmware family (4.80, NVM 1.3805.0, DDP 1.3.43.0) using four PFs in vlan-filtering bridges with vid 2-4094 and the slaves brought to IFF_PROMISC before the bridge VLAN bulk add:

before fix: ~12,279 AQ Add Switch Rules per PF, ENOSPC and ENOENT
responses in dmesg, DFLT VSI Rx rule install fails on the affected PF
after fix: ~4,093 AQ Add Switch Rules per PF, no AQ errors, DFLT
VSI Rx rule installs on every PF

The 66.7% reduction in installed switch rules per PF matches the expected per-VLAN saving: a single DFLT rule replaces the per-VID PROMISC_VLAN expansion.

Functional regression test with vid 2-100 trunk between two ice ports through the lab switch (40/40 PASS, 0 AQ errors, 0 ENOSPC at 4093-VID customer scale):

vid 50 unicast, vid 100 unicast, vid 50 broadcast ARP,
vid 100 multicast IPv6 ND
vid 200/500/1500/4000 isolation (out-of-trunk) and untagged not
leaked: 0 packets reach any bridge endpoint
IGMP/MLD snooping, Jumbo MTU 9000, reserved-multicast STP BPDU IFF_PROMISC + IFF_ALLMULTI transition (off while allmulti stays) Regression reproducer for commit 1273f89578f2 ("ice: Fix broken

    IFF_ALLMULTI handling"): allmulti on -> add vid -> allmulti off
    -> allmulti on plus the orphan-rule Scenario 2; both converge
    with no stale rules

100-VID, 1000-VID, 4093-VID stress cycles (5/3/2 iterations each) switchdev mode toggle preserves IFF_PROMISC pruning state across
the session (vid 999 multicast received before and after the legacy -> switchdev -> legacy cycle)
SR-IOV: VFs unaffected because ice_set_promisc() early-returns
for non-PF VSI and VF representors do not register ndo_vlan_rx_add_vid

Fixes 1273f89578f2 ("ice: Fix broken IFF_ALLMULTI handling")
---
 drivers/net/ethernet/intel/ice/ice_main.c | 90 ++++++++++++++++++-----
 1 file changed, 70 insertions(+), 20 deletions(-)
drivers/net/ethernet/intel/ice/ice_main.c +70 -20
index 6d24056c247cf4..af8df81fc45623 100644
@@ -274,7 +274,8 @@ static int ice_set_promisc(struct ice_vsi *vsi, u8 promisc_m)
 	if (vsi->type != ICE_VSI_PF)
 		return 0;
 
-	if (ice_vsi_has_non_zero_vlans(vsi)) {
+	/* skip per-VID expansion; the DFLT Rx rule already covers every VID */
+	if (ice_vsi_has_non_zero_vlans(vsi) && !ice_is_vsi_dflt_vsi(vsi)) {
 		promisc_m |= (ICE_PROMISC_VLAN_RX | ICE_PROMISC_VLAN_TX);
 		status = ice_fltr_set_vlan_vsi_promisc(&vsi->back->hw, vsi,
 						       promisc_m);
@@ -304,9 +305,19 @@ static int ice_clear_promisc(struct ice_vsi *vsi, u8 promisc_m)
 		return 0;
 
 	if (ice_vsi_has_non_zero_vlans(vsi)) {
-		promisc_m |= (ICE_PROMISC_VLAN_RX | ICE_PROMISC_VLAN_TX);
+		int vid0_status;
+
+		/* set time used either recipe (per-VID PROMISC_VLAN, or vid=0
+		 * PROMISC via the ice_set_promisc() else branch), so clear
+		 * both; clearing an absent rule succeeds
+		 */
 		status = ice_fltr_clear_vlan_vsi_promisc(&vsi->back->hw, vsi,
-							 promisc_m);
+				promisc_m | ICE_PROMISC_VLAN_RX |
+				ICE_PROMISC_VLAN_TX);
+		vid0_status = ice_fltr_clear_vsi_promisc(&vsi->back->hw,
+							 vsi->idx, promisc_m, 0);
+		if (!status)
+			status = vid0_status;
 	} else {
 		status = ice_fltr_clear_vsi_promisc(&vsi->back->hw, vsi->idx,
 						    promisc_m, 0);
@@ -317,6 +328,49 @@ static int ice_clear_promisc(struct ice_vsi *vsi, u8 promisc_m)
 	return status;
 }
 
+/**
+ * ice_vsi_exit_dflt_promisc - drop the default VSI Rx rule on promisc off
+ * @vsi: the VSI leaving promiscuous mode
+ *
+ * For an IFF_ALLMULTI VSI with VLANs the per-VID multicast rules are
+ * reinstated before the default rule is cleared so coverage never lapses;
+ * the then redundant vid=0 rule is dropped best-effort. The callees log
+ * their own failures, so error returns are not re-logged here.
+ *
+ * Return: 0 on success, negative on error with the default rule left in place.
+ */
+static int ice_vsi_exit_dflt_promisc(struct ice_vsi *vsi)
+{
+	struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi);
+	struct net_device *netdev = vsi->netdev;
+	struct ice_hw *hw = &vsi->back->hw;
+	bool restore_mc;
+	int err;
+
+	restore_mc = (vsi->current_netdev_flags & IFF_ALLMULTI) &&
+		     ice_vsi_has_non_zero_vlans(vsi);
+
+	if (restore_mc) {
+		err = ice_fltr_set_vlan_vsi_promisc(hw, vsi,
+						    ICE_MCAST_VLAN_PROMISC_BITS);
+		if (err && err != -EEXIST)
+			return err;
+	}
+
+	err = ice_clear_dflt_vsi(vsi);
+	if (err)
+		return err;
+
+	if (netdev->features & NETIF_F_HW_VLAN_CTAG_FILTER)
+		vlan_ops->ena_rx_filtering(vsi);
+
+	if (restore_mc)
+		ice_fltr_clear_vsi_promisc(hw, vsi->idx, ICE_MCAST_PROMISC_BITS,
+					   0);
+
+	return 0;
+}
+
 /**
  * ice_vsi_sync_fltr - Update the VSI filter list to the HW
  * @vsi: ptr to the VSI
@@ -442,17 +496,12 @@ static int ice_vsi_sync_fltr(struct ice_vsi *vsi)
 		} else {
 			/* Clear Rx filter to remove traffic from wire */
 			if (ice_is_vsi_dflt_vsi(vsi)) {
-				err = ice_clear_dflt_vsi(vsi);
+				err = ice_vsi_exit_dflt_promisc(vsi);
 				if (err) {
-					netdev_err(netdev, "Error %d clearing default VSI %i Rx rule\n",
-						   err, vsi->vsi_num);
 					vsi->current_netdev_flags |=
 						IFF_PROMISC;
 					goto out_promisc;
 				}
-				if (vsi->netdev->features &
-				    NETIF_F_HW_VLAN_CTAG_FILTER)
-					vlan_ops->ena_rx_filtering(vsi);
 			}
 
 			/* disable allmulti here, but only if allmulti is not
@@ -3676,10 +3725,9 @@ int ice_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
 	while (test_and_set_bit(ICE_CFG_BUSY, vsi->state))
 		usleep_range(1000, 2000);
 
-	/* Add multicast promisc rule for the VLAN ID to be added if
-	 * all-multicast is currently enabled.
-	 */
-	if (vsi->current_netdev_flags & IFF_ALLMULTI) {
+	/* skip the per-VID rule when the DFLT Rx rule already covers this VID */
+	if ((vsi->current_netdev_flags & IFF_ALLMULTI) &&
+	    !ice_is_vsi_dflt_vsi(vsi)) {
 		ret = ice_fltr_set_vsi_promisc(&vsi->back->hw, vsi->idx,
 					       ICE_MCAST_VLAN_PROMISC_BITS,
 					       vid);
@@ -3697,11 +3745,12 @@ int ice_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
 	if (ret)
 		goto finish;
 
-	/* If all-multicast is currently enabled and this VLAN ID is only one
-	 * besides VLAN-0 we have to update look-up type of multicast promisc
-	 * rule for VLAN-0 from ICE_SW_LKUP_PROMISC to ICE_SW_LKUP_PROMISC_VLAN.
+	/* On the first non-zero VLAN, promote the VLAN-0 multicast promisc
+	 * rule from ICE_SW_LKUP_PROMISC to ICE_SW_LKUP_PROMISC_VLAN. Skip when
+	 * the DFLT Rx rule is installed; it already covers every VID.
 	 */
 	if ((vsi->current_netdev_flags & IFF_ALLMULTI) &&
+	    !ice_is_vsi_dflt_vsi(vsi) &&
 	    ice_vsi_num_non_zero_vlans(vsi) == 1) {
 		ice_fltr_clear_vsi_promisc(&vsi->back->hw, vsi->idx,
 					   ICE_MCAST_PROMISC_BITS, 0);
@@ -3764,11 +3813,12 @@ int ice_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
 					   ICE_MCAST_VLAN_PROMISC_BITS, vid);
 
 	if (!ice_vsi_has_non_zero_vlans(vsi)) {
-		/* Update look-up type of multicast promisc rule for VLAN 0
-		 * from ICE_SW_LKUP_PROMISC_VLAN to ICE_SW_LKUP_PROMISC when
-		 * all-multicast is enabled and VLAN 0 is the only VLAN rule.
+		/* Last non-zero VLAN gone: demote the VLAN-0 multicast promisc
+		 * rule back to ICE_SW_LKUP_PROMISC. Skip when the DFLT Rx rule
+		 * is installed; no recipe swap is needed.
 		 */
-		if (vsi->current_netdev_flags & IFF_ALLMULTI) {
+		if ((vsi->current_netdev_flags & IFF_ALLMULTI) &&
+		    !ice_is_vsi_dflt_vsi(vsi)) {
 			ice_fltr_clear_vsi_promisc(&vsi->back->hw, vsi->idx,
 						   ICE_MCAST_VLAN_PROMISC_BITS,
 						   0);
Signature · 1 line
2.53.0
author
#3 lore raw

PATCH 2/2 ice: preserve uplink DFLT Rx rule on switchdev release

In reply to Petr Oros

ice_eswitch_setup_env() calls ice_set_dflt_vsi() to install the ICE_SW_LKUP_DFLT Rx rule on the uplink VSI. The helper returns 0 even when the rule is already in place, so the call is a no-op if ice_vsi_sync_fltr() had previously installed the DFLT rule in response to IFF_PROMISC on the uplink netdev. ice_remove_vsi_fltr() called earlier in ice_eswitch_setup_env() does not affect this rule because ice_remove_vsi_lkup_fltr() lacks a case for ICE_SW_LKUP_DFLT and falls into its default branch which only logs. Switchdev mode then adds an ICE_FLTR_TX leg via ice_cfg_dflt_vsi() on the same VSI handle.

ice_eswitch_release_env() unconditionally removed both the Rx and Tx DFLT rules. When the Rx DFLT was installed by ice_vsi_sync_fltr() before the switchdev session started, this clobbered promisc state the operator had asked for: the DFLT Rx rule disappeared while IFF_PROMISC was still set on the netdev, and the IFF_PROMISC sync path was not retriggered, so the uplink ended the session without the catch-all rule the netdev flags requested.

Skip the Rx DFLT removal when the uplink is still promiscuous, both in ice_eswitch_release_env() and in the err_def_tx unwind of ice_eswitch_setup_env(). The Tx leg installed by switchdev is always removed since switchdev owns it.

The ena_rx_filtering() call earlier in ice_eswitch_release_env() is left unconditional: it calls ice_cfg_vlan_pruning(), which returns without enabling pruning while the netdev is in IFF_PROMISC, so it cannot re-enable VLAN pruning under the preserved DFLT rule and drop tagged traffic. Pruning is re-enabled later, when the IFF_PROMISC sync path runs after promisc is actually cleared.

Use vsi->current_netdev_flags rather than the live netdev->flags for this test. netdev->flags is written under RTNL by dev_change_flags(), while ice_eswitch_release_env() runs under devl_lock, so reading it here would be a TOCTOU against a concurrent promisc change. The IFF_PROMISC bit of current_netdev_flags is written only under ICE_CFG_BUSY by ice_vsi_sync_fltr(), and ice_set_rx_mode() gates that sync off for the uplink while ice_is_switchdev_running() is true. The bit is therefore frozen for the whole session and stable when release_env reads it.

Because the sync is gated off during the session, a promisc change the operator makes while switchdev runs never reaches ice_vsi_sync_fltr():
current_netdev_flags keeps the value captured before the session while netdev->flags carries the new one. Once switchdev is torn down and pf->eswitch.is_running is cleared, schedule a filter sync from ice_eswitch_disable_switchdev() so the suppressed change is replayed and the DFLT Rx rule is reconciled with the current netdev flags. This also closes the window where release_env kept the rule based on the frozen flag but the operator had since cleared IFF_PROMISC.

Fixes 1a1c40df2e80 ("ice: set and release switchdev environment")
---
 drivers/net/ethernet/intel/ice/ice_eswitch.c | 32 +++++++++++++++++---
 1 file changed, 28 insertions(+), 4 deletions(-)
drivers/net/ethernet/intel/ice/ice_eswitch.c +28 -4
index 2e4f0969035f77..b6073fc2375019 100644
@@ -66,8 +66,10 @@ static int ice_eswitch_setup_env(struct ice_pf *pf)
 	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
 			 ICE_FLTR_TX);
 err_def_tx:
-	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
-			 ICE_FLTR_RX);
+	/* keep the Rx DFLT rule if still promiscuous (see release_env) */
+	if (!(uplink_vsi->current_netdev_flags & IFF_PROMISC))
+		ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx,
+				 false, ICE_FLTR_RX);
 err_def_rx:
 	ice_vsi_del_vlan_zero(uplink_vsi);
 err_vlan_zero:
@@ -275,11 +277,23 @@ static void ice_eswitch_release_env(struct ice_pf *pf)
 	vlan_ops = ice_get_compat_vsi_vlan_ops(uplink_vsi);
 
 	ice_vsi_update_local_lb(uplink_vsi, false);
+	/* No-op while IFF_PROMISC is set: ice_cfg_vlan_pruning() self-gates on
+	 * it, so this cannot re-enable VLAN pruning under a preserved DFLT rule.
+	 */
 	vlan_ops->ena_rx_filtering(uplink_vsi);
 	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
 			 ICE_FLTR_TX);
-	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
-			 ICE_FLTR_RX);
+
+	/* Keep the Rx DFLT rule if the uplink is still promiscuous; it must
+	 * outlive the session. current_netdev_flags is used because its
+	 * IFF_PROMISC bit only changes under ice_vsi_sync_fltr(), gated off
+	 * during switchdev, so the read cannot race the RTNL netdev->flags.
+	 * Any change made during the session is replayed on teardown.
+	 */
+	if (!(uplink_vsi->current_netdev_flags & IFF_PROMISC))
+		ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx,
+				 false, ICE_FLTR_RX);
+
 	ice_fltr_add_mac_and_broadcast(uplink_vsi,
 				       uplink_vsi->port_info->mac.perm_addr,
 				       ICE_FWD_TO_VSI);
@@ -327,10 +341,20 @@ static int ice_eswitch_enable_switchdev(struct ice_pf *pf)
  */
 static void ice_eswitch_disable_switchdev(struct ice_pf *pf)
 {
+	struct ice_vsi *uplink_vsi = pf->eswitch.uplink_vsi;
+
 	ice_eswitch_br_offloads_deinit(pf);
 	ice_eswitch_release_env(pf);
 
 	pf->eswitch.is_running = false;
+
+	/* ice_set_rx_mode() was gated off during the session; replay a filter
+	 * sync so any suppressed promisc change reconciles the DFLT Rx rule.
+	 */
+	set_bit(ICE_VSI_UMAC_FLTR_CHANGED, uplink_vsi->state);
+	set_bit(ICE_VSI_MMAC_FLTR_CHANGED, uplink_vsi->state);
+	set_bit(ICE_FLAG_FLTR_SYNC, pf->flags);
+	ice_service_task_schedule(pf);
 }
 
 /**
Signature · 1 line
2.53.0

Petr Oros

author
#0 lore raw

cover ice: fix DFLT Rx rule handling for promisc and switchdev

Two fixes for the uplink default VSI Rx rule (DFLT) on E810 when the netdev is in IFF_PROMISC.

Patch 1 drops the redundant per-VLAN promisc expansion that exhausts the FLU pool on a wide VLAN trunk across several PFs.

Patch 2 keeps the DFLT Rx rule across a switchdev teardown instead of clobbering the promisc state the operator asked for.

Lab tested on E810-C: functional, VLAN isolation, IFF_ALLMULTI regression, stress/flap and switchdev-toggle suites pass with no AQ errors, and the FLU pool stays under its ceiling with all four PFs loaded.

Petr Oros (2):
ice: skip per-VLAN promisc rules when default VSI Rx rule is set
ice: preserve uplink DFLT Rx rule on switchdev release

 drivers/net/ethernet/intel/ice/ice_eswitch.c | 32 ++++++-
 drivers/net/ethernet/intel/ice/ice_main.c    | 90 +++++++++++++++-----
 2 files changed, 98 insertions(+), 24 deletions(-)
Signature · 1 line
2.53.0
Reviewed-by on v1 1/2
#2 lore raw

RE: [Intel-wired-lan] [PATCH iwl-net 1/2] ice: skip per-VLAN promisc rules when default VSI Rx rule is set

In reply to Petr Oros
Quoted 195 lines

-----Original Message-----
From: Intel-wired-lan <[email protected]> On Behalf Of Petr Oros
Sent: Thursday, June 18, 2026 5:09 PM
To: [email protected]
Cc: Vecera, Ivan <[email protected]>; Alice Michael <[email protected]>; Kitszel, Przemyslaw <[email protected]>; Eric Dumazet <[email protected]>; [email protected]; Andrew Lunn <[email protected]>; Nguyen, Anthony L <[email protected]>; Michal Swiatkowski <[email protected]>; Keller, Jacob E <[email protected]>; Jakub Kicinski <[email protected]>; Paolo Abeni <[email protected]>; David S. Miller <[email protected]>; [email protected]
Subject: [Intel-wired-lan] [PATCH iwl-net 1/2] ice: skip per-VLAN promisc rules when default VSI Rx rule is set

When an ice port is part of a vlan-filtering bridge with a wide VLAN trunk and the netdev is in IFF_PROMISC (typical for bond slaves attached to a bridge), the driver installs per-VLAN ICE_SW_LKUP_PROMISC_VLAN entries (recipe 9) in addition to the broad ICE_SW_LKUP_DFLT VSI Rx rule (recipe 5). Each per-VLAN rule consumes one Flow Lookup Unit (FLU) entry from a fixed hardware pool of "up to 32K FLU entries" per device, documented in the E810 datasheet (613875-009 section 7.8.10, Table 7-18, page 1015).

With three active PFs sharing one switch context and a bridge trunk of vid 2-4094, the configuration would require roughly

3 PFs * 4093 VLANs * 3 rules per VLAN per PF ~= 36,800 rules

which exceeds the 32K FLU budget. Firmware then responds to further Add Switch Rules requests with AQ retval 0x10 (LIBIE_AQ_RC_ENOSPC) and the user-visible failure surfaces as

ice 0000:5c:00.1: Failed to set VSI 14 as the default forwarding
VSI, error -5
ice 0000:5c:00.1 ens1f1: Error -5 setting default VSI 14 Rx rule

After a switch context has been driven into overrun, subsequent retries can come back as AQ retval 0x2 (LIBIE_AQ_RC_ENOENT), which has misled triage attempts toward a perceived recipe binding defect rather than a capacity issue.

When the DFLT VSI Rx rule is in place it catches every packet on the lport regardless of VLAN tag, so the per-VLAN PROMISC_VLAN expansion is redundant. The recipe 4 VLAN prune entries are still installed per VLAN and continue to track the allowed VID set, but the IFF_PROMISC sync path disables their enforcement on the VSI via vlan_ops->dis_rx_filtering() before ice_set_promisc() runs. ena_rx_filtering() is restored when IFF_PROMISC is cleared.

Skip the per-VLAN expansion at the two call sites that drive it:
ice_set_promisc() falls through to ice_fltr_set_vsi_promisc() and ice_vlan_rx_add_vid() omits the per-VLAN ICE_MCAST_VLAN_PROMISC_BITS add. Plain IFF_ALLMULTI without an installed DFLT VSI rule is unchanged and still installs per-VLAN multicast promisc rules.

Both checks use ice_is_vsi_dflt_vsi() which inspects the recipe filter list for an installed DFLT rule on this VSI, not netdev->flags & IFF_PROMISC. The HW-state predicate avoids two regression vectors that a user-intent predicate would introduce:

1. ice_lag_is_switchdev_running() short-circuits ice_set_dflt_vsi() to return 0 without installing the DFLT rule for a PF in switchdev LAG mode. An IFF_PROMISC-only check would also suppress the per-VLAN fallback, leaving the PF with no rule.

2. When ice_set_dflt_vsi() returns a non-EEXIST error (FLU exhausted, switch context divergence), the driver clears IFF_PROMISC from vsi->current_netdev_flags but the netdev's own flags retain IFF_PROMISC. The user-intent predicate would still suppress the per-VLAN fallback even though DFLT failed to install.

The predicate is install-time only. The IFF_PROMISC off path closes the lifecycle gap in ice_vsi_exit_dflt_promisc(): for an IFF_ALLMULTI VSI with VLANs it reinstates the per-VID rules before clearing the default rule, so multicast coverage never lapses. If that AQ call fails the default rule is left in place, ice_vsi_exit_dflt_promisc() returns the error, and the sync_fltr pass bails with vsi->current_netdev_flags |= IFF_PROMISC; the current/netdev flag mismatch re-fires the IFF_PROMISC off path on the next sync. Clearing the default rule first would instead expose a window where neither the default rule nor the per-VID rules carry multicast.

If ice_clear_dflt_vsi() fails after the per-VID rules were reinstated they are deliberately not rolled back. Clearing the default rule is a removal that frees an FLU entry rather than allocating one, so it cannot fail for lack of space; a failure is a transient AdminQ error. The per-VID rules are the steady state for an IFF_ALLMULTI VLAN VSI, so the only redundant entry left behind is the single un-removed default rule, not the per-VID set. The retry re-enters this path, ice_fltr_set_vlan_vsi_promisc() returns -EEXIST for the rules that already exist so nothing is reallocated, and the default rule is removed on the next attempt. Rolling the per-VID rules back here would instead churn thousands of removes and re-adds on every retry.

After the default rule is gone the vid=0 PROMISC rule that paired with it is redundant and is dropped, but only to reclaim a filter entry, so a failure there is logged and does not abort the transition.

ice_set_vsi_promisc() and ice_clear_vsi_promisc() dispatch the recipe based on whether ICE_PROMISC_VLAN_RX/TX bits are present in the mask:
with the bits set, recipe ICE_SW_LKUP_PROMISC_VLAN is used; otherwise ICE_SW_LKUP_PROMISC. The else branch in
ice_set_promisc() installs the vid=0 rule in ICE_SW_LKUP_PROMISC. Because ice_clear_promisc() with VLANs present adds the VLAN bits and would search ICE_SW_LKUP_PROMISC_VLAN, the recipe mismatch would leave the vid=0 ICE_SW_LKUP_PROMISC rule orphaned when VLANs are configured. This is a single stale rule, not a per-cycle leak:
re-adding it on the next promisc on returns -EEXIST rather than allocating a new entry. The set-time recipe is not recorded, so ice_clear_promisc() clears both recipes; clearing a rule that is not present succeeds, both clears run unconditionally, and the first error is returned.

The two VLAN-0 recipe transition blocks in ice_vlan_rx_add_vid() and ice_vlan_rx_kill_vid() that promote / demote the vid=0 rule between ICE_SW_LKUP_PROMISC and ICE_SW_LKUP_PROMISC_VLAN are likewise guarded by !ice_is_vsi_dflt_vsi(). With DFLT in place the
vid=0 rule already covers every VID and a recipe swap would only install a redundant rule.

Lab reproduction on an E810-C with the same firmware family (4.80, NVM 1.3805.0, DDP 1.3.43.0) using four PFs in vlan-filtering bridges with vid 2-4094 and the slaves brought to IFF_PROMISC before the bridge VLAN bulk add:

before fix: ~12,279 AQ Add Switch Rules per PF, ENOSPC and ENOENT
responses in dmesg, DFLT VSI Rx rule install fails on the affected PF
after fix: ~4,093 AQ Add Switch Rules per PF, no AQ errors, DFLT
VSI Rx rule installs on every PF

The 66.7% reduction in installed switch rules per PF matches the expected per-VLAN saving: a single DFLT rule replaces the per-VID PROMISC_VLAN expansion.

Functional regression test with vid 2-100 trunk between two ice ports through the lab switch (40/40 PASS, 0 AQ errors, 0 ENOSPC at 4093-VID customer scale):

vid 50 unicast, vid 100 unicast, vid 50 broadcast ARP,
vid 100 multicast IPv6 ND
vid 200/500/1500/4000 isolation (out-of-trunk) and untagged not
leaked: 0 packets reach any bridge endpoint
IGMP/MLD snooping, Jumbo MTU 9000, reserved-multicast STP BPDU IFF_PROMISC + IFF_ALLMULTI transition (off while allmulti stays) Regression reproducer for commit 1273f89578f2 ("ice: Fix broken

    IFF_ALLMULTI handling"): allmulti on -> add vid -> allmulti off
    -> allmulti on plus the orphan-rule Scenario 2; both converge
    with no stale rules

100-VID, 1000-VID, 4093-VID stress cycles (5/3/2 iterations each) switchdev mode toggle preserves IFF_PROMISC pruning state across
the session (vid 999 multicast received before and after the legacy -> switchdev -> legacy cycle)
SR-IOV: VFs unaffected because ice_set_promisc() early-returns
for non-PF VSI and VF representors do not register ndo_vlan_rx_add_vid

Fixes 1273f89578f2 ("ice: Fix broken IFF_ALLMULTI handling")
---
 drivers/net/ethernet/intel/ice/ice_main.c | 90 ++++++++++++++++++----
(quoted diff) +0 -1
-
 1 file changed, 70 insertions(+), 20 deletions(-)
drivers/net/ethernet/intel/ice/ice_main.c +1 -1
index 6d24056c247cf4..af8df81fc45623 100644
@@ -274,7 +274,8 @@ static int ice_set_promisc(struct ice_vsi *vsi, u8
 promisc_m)
 	if (vsi->type != ICE_VSI_PF)
 		return 0;
 
-	if (ice_vsi_has_non_zero_vlans(vsi)) {
+	/* skip per-VID expansion; the DFLT Rx rule already covers

every VID */

drivers/net/ethernet/intel/ice/ice_main.c +1 -0
+	if (ice_vsi_has_non_zero_vlans(vsi) &&

!ice_is_vsi_dflt_vsi(vsi)) {
promisc_m |= (ICE_PROMISC_VLAN_RX |
ICE_PROMISC_VLAN_TX);
status = ice_fltr_set_vlan_vsi_promisc(&vsi->back->hw,
vsi,

drivers/net/ethernet/intel/ice/ice_main.c +0 -0
 						       promisc_m);
@@ -304,9 +305,19 @@ static int ice_clear_promisc(struct ice_vsi *vsi,
 u8 promisc_m)
 		return 0;
 
 	if (ice_vsi_has_non_zero_vlans(vsi)) {

...

Quoted 7 lines

ice_fltr_clear_vsi_promisc(&vsi->back->hw, vsi-

Quoted 1 line

idx,

ICE_MCAST_VLAN_PROMISC_BITS,
0);
--
2.53.0

Reviewed-by on v1 2/2
#4 lore raw

RE: [Intel-wired-lan] [PATCH iwl-net 2/2] ice: preserve uplink DFLT Rx rule on switchdev release

In reply to Petr Oros
Quoted 153 lines

-----Original Message-----
From: Intel-wired-lan <[email protected]> On Behalf Of Petr Oros
Sent: Thursday, June 18, 2026 5:09 PM
To: [email protected]
Cc: Vecera, Ivan <[email protected]>; Alice Michael <[email protected]>; Kitszel, Przemyslaw <[email protected]>; Eric Dumazet <[email protected]>; [email protected]; Andrew Lunn <[email protected]>; Nguyen, Anthony L <[email protected]>; Michal Swiatkowski <[email protected]>; Keller, Jacob E <[email protected]>; Jakub Kicinski <[email protected]>; Paolo Abeni <[email protected]>; David S. Miller <[email protected]>; [email protected]
Subject: [Intel-wired-lan] [PATCH iwl-net 2/2] ice: preserve uplink DFLT Rx rule on switchdev release

ice_eswitch_setup_env() calls ice_set_dflt_vsi() to install the ICE_SW_LKUP_DFLT Rx rule on the uplink VSI. The helper returns 0 even when the rule is already in place, so the call is a no-op if ice_vsi_sync_fltr() had previously installed the DFLT rule in response to IFF_PROMISC on the uplink netdev. ice_remove_vsi_fltr() called earlier in ice_eswitch_setup_env() does not affect this rule because ice_remove_vsi_lkup_fltr() lacks a case for ICE_SW_LKUP_DFLT and falls into its default branch which only logs. Switchdev mode then adds an ICE_FLTR_TX leg via ice_cfg_dflt_vsi() on the same VSI handle.

ice_eswitch_release_env() unconditionally removed both the Rx and Tx DFLT rules. When the Rx DFLT was installed by ice_vsi_sync_fltr() before the switchdev session started, this clobbered promisc state the operator had asked for: the DFLT Rx rule disappeared while IFF_PROMISC was still set on the netdev, and the IFF_PROMISC sync path was not retriggered, so the uplink ended the session without the catch-all rule the netdev flags requested.

Skip the Rx DFLT removal when the uplink is still promiscuous, both in ice_eswitch_release_env() and in the err_def_tx unwind of ice_eswitch_setup_env(). The Tx leg installed by switchdev is always removed since switchdev owns it.

The ena_rx_filtering() call earlier in ice_eswitch_release_env() is left unconditional: it calls ice_cfg_vlan_pruning(), which returns without enabling pruning while the netdev is in IFF_PROMISC, so it cannot re-enable VLAN pruning under the preserved DFLT rule and drop tagged traffic. Pruning is re-enabled later, when the IFF_PROMISC sync path runs after promisc is actually cleared.

Use vsi->current_netdev_flags rather than the live netdev->flags for this test. netdev->flags is written under RTNL by dev_change_flags(), while ice_eswitch_release_env() runs under devl_lock, so reading it here would be a TOCTOU against a concurrent promisc change. The IFF_PROMISC bit of current_netdev_flags is written only under ICE_CFG_BUSY by ice_vsi_sync_fltr(), and ice_set_rx_mode() gates that sync off for the uplink while ice_is_switchdev_running() is true. The bit is therefore frozen for the whole session and stable when release_env reads it.

Because the sync is gated off during the session, a promisc change the operator makes while switchdev runs never reaches ice_vsi_sync_fltr():
current_netdev_flags keeps the value captured before the session while netdev->flags carries the new one. Once switchdev is torn down and pf->eswitch.is_running is cleared, schedule a filter sync from ice_eswitch_disable_switchdev() so the suppressed change is replayed and the DFLT Rx rule is reconciled with the current netdev flags. This also closes the window where release_env kept the rule based on the frozen flag but the operator had since cleared IFF_PROMISC.

Fixes 1a1c40df2e80 ("ice: set and release switchdev environment")
---
 drivers/net/ethernet/intel/ice/ice_eswitch.c | 32 +++++++++++++++++--
(quoted diff) +0 -1
-
 1 file changed, 28 insertions(+), 4 deletions(-)
drivers/net/ethernet/intel/ice/ice_eswitch.c +4 -2
index 2e4f0969035f77..b6073fc2375019 100644
@@ -66,8 +66,10 @@ static int ice_eswitch_setup_env(struct ice_pf *pf)
 	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
 			 ICE_FLTR_TX);
 err_def_tx:
-	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
-			 ICE_FLTR_RX);
+	/* keep the Rx DFLT rule if still promiscuous (see release_env)
 */
+	if (!(uplink_vsi->current_netdev_flags & IFF_PROMISC))
+		ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx,
+				 false, ICE_FLTR_RX);
 err_def_rx:
 	ice_vsi_del_vlan_zero(uplink_vsi);

err_vlan_zero:

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
@@ -275,11 +277,23 @@ static void ice_eswitch_release_env(struct
 ice_pf *pf)
 	vlan_ops = ice_get_compat_vsi_vlan_ops(uplink_vsi);
 
 	ice_vsi_update_local_lb(uplink_vsi, false);
+	/* No-op while IFF_PROMISC is set: ice_cfg_vlan_pruning() self-

gates on

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
+	 * it, so this cannot re-enable VLAN pruning under a preserved

DFLT rule.

drivers/net/ethernet/intel/ice/ice_eswitch.c +3 -2
+	 */
 	vlan_ops->ena_rx_filtering(uplink_vsi);
 	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
 			 ICE_FLTR_TX);
-	ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
-			 ICE_FLTR_RX);
+
+	/* Keep the Rx DFLT rule if the uplink is still promiscuous; it

must

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
+	 * outlive the session. current_netdev_flags is used because

its

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
+	 * IFF_PROMISC bit only changes under ice_vsi_sync_fltr(),

gated off

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
+	 * during switchdev, so the read cannot race the RTNL netdev-
Quoted 1 line

flags.

drivers/net/ethernet/intel/ice/ice_eswitch.c +6 -0
+	 * Any change made during the session is replayed on teardown.
+	 */
+	if (!(uplink_vsi->current_netdev_flags & IFF_PROMISC))
+		ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx,
+				 false, ICE_FLTR_RX);
+
 	ice_fltr_add_mac_and_broadcast(uplink_vsi,
 				       uplink_vsi->port_info-
Quoted 1 line

mac.perm_addr,

drivers/net/ethernet/intel/ice/ice_eswitch.c +4 -0
 				       ICE_FWD_TO_VSI);
@@ -327,10 +341,20 @@ static int ice_eswitch_enable_switchdev(struct
 ice_pf *pf)
  */
 static void ice_eswitch_disable_switchdev(struct ice_pf *pf)  {
+	struct ice_vsi *uplink_vsi = pf->eswitch.uplink_vsi;
+
 	ice_eswitch_br_offloads_deinit(pf);
 	ice_eswitch_release_env(pf);
 
 	pf->eswitch.is_running = false;
+
+	/* ice_set_rx_mode() was gated off during the session; replay a

filter

drivers/net/ethernet/intel/ice/ice_eswitch.c +1 -0
+	 * sync so any suppressed promisc change reconciles the DFLT Rx

rule.

drivers/net/ethernet/intel/ice/ice_eswitch.c +5 -1
+	 */
+	set_bit(ICE_VSI_UMAC_FLTR_CHANGED, uplink_vsi->state);
+	set_bit(ICE_VSI_MMAC_FLTR_CHANGED, uplink_vsi->state);
+	set_bit(ICE_FLAG_FLTR_SYNC, pf->flags);
+	ice_service_task_schedule(pf);
 }
 
 /**
--
Quoted 1 line

2.53.0