Merge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/gi…

…t/tnguy/next-queue

Tony Nguyen says:

====================
ice: use less resources in switchdev

Michal Swiatkowski says:

Switchdev is using one queue per created port representor. This can
quickly lead to Rx queue shortage, as with subfunction support user
can create high number of PRs.

Save one MSI-X and 'number of PRs' * 1 queues.
Refactor switchdev slow-path to use less resources (even no additional
resources). Do this by removing control plane VSI and move its
functionality to PF VSI. Even with current solution PF is acting like
uplink and can't be used simultaneously for other use cases (adding
filters can break slow-path).

In short, do Tx via PF VSI and Rx packets using PF resources. Rx needs
additional code in interrupt handler to choose correct PR netdev.
Previous solution had to queue filters, it was way more elegant but
needed one queue per PRs. Beside that this refactor mostly simplifies
switchdev configuration.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  ice: count representor stats
  ice: do switchdev slow-path Rx using PF VSI
  ice: change repr::id values
  ice: remove switchdev control plane VSI
  ice: control default Tx rule in lag
  ice: default Tx rule instead of to queue
  ice: do Tx through PF netdev in slow-path
  ice: remove eswitch changing queues algorithm
====================

Link: https://lore.kernel.org/r/20240325202623.1012287-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/intel/ice/ice.h

@@ -522,17 +522,10 @@ enum ice_misc_thread_tasks {
 };
 
 struct ice_eswitch {
-	struct ice_vsi *control_vsi;
 	struct ice_vsi *uplink_vsi;
 	struct ice_esw_br_offloads *br_offloads;
 	struct xarray reprs;
 	bool is_running;
-	/* struct to allow cp queues management optimization */
-	struct {
-		int to_reach;
-		int value;
-		bool is_reaching;
-	} qs;
 };
 
 struct ice_agg_node {

drivers/net/ethernet/intel/ice/ice_base.c

@@ -263,30 +263,6 @@ static u16 ice_calc_txq_handle(struct ice_vsi *vsi, struct ice_tx_ring *ring, u8
 	return ring->q_index - vsi->tc_cfg.tc_info[tc].qoffset;
 }
 
-/**
- * ice_eswitch_calc_txq_handle
- * @ring: pointer to ring which unique index is needed
- *
- * To correctly work with many netdevs ring->q_index of Tx rings on switchdev
- * VSI can repeat. Hardware ring setup requires unique q_index. Calculate it
- * here by finding index in vsi->tx_rings of this ring.
- *
- * Return ICE_INVAL_Q_INDEX when index wasn't found. Should never happen,
- * because VSI is get from ring->vsi, so it has to be present in this VSI.
- */
-static u16 ice_eswitch_calc_txq_handle(struct ice_tx_ring *ring)
-{
-	const struct ice_vsi *vsi = ring->vsi;
-	int i;
-
-	ice_for_each_txq(vsi, i) {
-		if (vsi->tx_rings[i] == ring)
-			return i;
-	}
-
-	return ICE_INVAL_Q_INDEX;
-}
-
 /**
  * ice_cfg_xps_tx_ring - Configure XPS for a Tx ring
  * @ring: The Tx ring to configure
@@ -353,9 +329,6 @@ ice_setup_tx_ctx(struct ice_tx_ring *ring, struct ice_tlan_ctx *tlan_ctx, u16 pf
 		tlan_ctx->vmvf_num = hw->func_caps.vf_base_id + vsi->vf->vf_id;
 		tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_VF;
 		break;
-	case ICE_VSI_SWITCHDEV_CTRL:
-		tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_VMQ;
-		break;
 	default:
 		return;
 	}
@@ -479,6 +452,14 @@ static int ice_setup_rx_ctx(struct ice_rx_ring *ring)
 	/* Rx queue threshold in units of 64 */
 	rlan_ctx.lrxqthresh = 1;
 
+	/* PF acts as uplink for switchdev; set flex descriptor with src_vsi
+	 * metadata and flags to allow redirecting to PR netdev
+	 */
+	if (ice_is_eswitch_mode_switchdev(vsi->back)) {
+		ring->flags |= ICE_RX_FLAGS_MULTIDEV;
+		rxdid = ICE_RXDID_FLEX_NIC_2;
+	}
+
 	/* Enable Flexible Descriptors in the queue context which
 	 * allows this driver to select a specific receive descriptor format
 	 * increasing context priority to pick up profile ID; default is 0x01;
@@ -919,14 +900,7 @@ ice_vsi_cfg_txq(struct ice_vsi *vsi, struct ice_tx_ring *ring,
 	/* Add unique software queue handle of the Tx queue per
 	 * TC into the VSI Tx ring
 	 */
-	if (vsi->type == ICE_VSI_SWITCHDEV_CTRL) {
-		ring->q_handle = ice_eswitch_calc_txq_handle(ring);
-
-		if (ring->q_handle == ICE_INVAL_Q_INDEX)
-			return -ENODEV;
-	} else {
-		ring->q_handle = ice_calc_txq_handle(vsi, ring, tc);
-	}
+	ring->q_handle = ice_calc_txq_handle(vsi, ring, tc);
 
 	if (ch)
 		status = ice_ena_vsi_txq(vsi->port_info, ch->ch_vsi->idx, 0,

drivers/net/ethernet/intel/ice/ice_dcb_lib.c

@@ -291,7 +291,6 @@ static void ice_dcb_ena_dis_vsi(struct ice_pf *pf, bool ena, bool locked)
 
 		switch (vsi->type) {
 		case ICE_VSI_CHNL:
-		case ICE_VSI_SWITCHDEV_CTRL:
 		case ICE_VSI_PF:
 			if (ena)
 				ice_ena_vsi(vsi, locked);
@@ -776,8 +775,7 @@ void ice_pf_dcb_recfg(struct ice_pf *pf, bool locked)
 		/* no need to proceed with remaining cfg if it is CHNL
 		 * or switchdev VSI
 		 */
-		if (vsi->type == ICE_VSI_CHNL ||
-		    vsi->type == ICE_VSI_SWITCHDEV_CTRL)
+		if (vsi->type == ICE_VSI_CHNL)
 			continue;
 
 		ice_vsi_map_rings_to_vectors(vsi);