diff mbox series

[iwl-next,v1,15/15] ice: reserve number of CP queues

Message ID 20231024110929.19423-16-michal.swiatkowski@linux.intel.com (mailing list archive)
State Awaiting Upstream
Delegated to: Netdev Maintainers
Headers show
Series one by one port representors creation | expand

Checks

Context Check Description
netdev/tree_selection success Guessing tree name failed - patch did not apply

Commit Message

Michal Swiatkowski Oct. 24, 2023, 11:09 a.m. UTC 
Rebuilding CP VSI each time the PR is created drastically increase the
time of maximum VFs creation. Add function to reserve number of CP
queues to deal with this problem.

Use the same function to decrease number of queues in case of removing
VFs. Assume that caller of ice_eswitch_reserve_cp_queues() will also
call ice_eswitch_attach/detach() correct number of times.

Still one by one PR adding is handy for VF resetting routine.

Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
---
 drivers/net/ethernet/intel/ice/ice.h         |  6 +++
 drivers/net/ethernet/intel/ice/ice_eswitch.c | 52 +++++++++++++++++---
 drivers/net/ethernet/intel/ice/ice_eswitch.h |  4 ++
 drivers/net/ethernet/intel/ice/ice_sriov.c   |  3 ++
 4 files changed, 58 insertions(+), 7 deletions(-)

Comments

Buvaneswaran, Sujai Nov. 9, 2023, 10:32 a.m. UTC | #1 
> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf Of
> Michal Swiatkowski
> Sent: Tuesday, October 24, 2023 4:39 PM
> To: intel-wired-lan@lists.osuosl.org
> Cc: Drewek, Wojciech <wojciech.drewek@intel.com>; Szycik, Marcin
> <marcin.szycik@intel.com>; netdev@vger.kernel.org; Brandeburg, Jesse
> <jesse.brandeburg@intel.com>; Kitszel, Przemyslaw
> <przemyslaw.kitszel@intel.com>; Keller, Jacob E <jacob.e.keller@intel.com>;
> Michal Swiatkowski <michal.swiatkowski@linux.intel.com>; Raczynski, Piotr
> <piotr.raczynski@intel.com>
> Subject: [Intel-wired-lan] [PATCH iwl-next v1 15/15] ice: reserve number of
> CP queues
> 
> Rebuilding CP VSI each time the PR is created drastically increase the time of
> maximum VFs creation. Add function to reserve number of CP queues to deal
> with this problem.
> 
> Use the same function to decrease number of queues in case of removing
> VFs. Assume that caller of ice_eswitch_reserve_cp_queues() will also call
> ice_eswitch_attach/detach() correct number of times.
> 
> Still one by one PR adding is handy for VF resetting routine.
> 
> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
> Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
> ---
>  drivers/net/ethernet/intel/ice/ice.h         |  6 +++
>  drivers/net/ethernet/intel/ice/ice_eswitch.c | 52 +++++++++++++++++---
> drivers/net/ethernet/intel/ice/ice_eswitch.h |  4 ++
>  drivers/net/ethernet/intel/ice/ice_sriov.c   |  3 ++
>  4 files changed, 58 insertions(+), 7 deletions(-)
> 
Tested-by: Sujai Buvaneswaran <sujai.buvaneswaran@intel.com>
diff mbox series

Patch

diff --git a/drivers/net/ethernet/intel/ice/ice.h b/drivers/net/ethernet/intel/ice/ice.h
index 597bdb6945c6..cd7dcd0fa7f2 100644
--- a/drivers/net/ethernet/intel/ice/ice.h
+++ b/drivers/net/ethernet/intel/ice/ice.h
@@ -528,6 +528,12 @@  struct ice_eswitch {
 	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 {
diff --git a/drivers/net/ethernet/intel/ice/ice_eswitch.c b/drivers/net/ethernet/intel/ice/ice_eswitch.c
index 9ff4fe4fb133..3f80e2081e5d 100644
--- a/drivers/net/ethernet/intel/ice/ice_eswitch.c
+++ b/drivers/net/ethernet/intel/ice/ice_eswitch.c
@@ -176,7 +176,7 @@  static void ice_eswitch_remap_rings_to_vectors(struct ice_eswitch *eswitch)
 
 		repr = xa_find(&eswitch->reprs, &repr_id, U32_MAX,
 			       XA_PRESENT);
-		if (WARN_ON(!repr))
+		if (!repr)
 			break;
 
 		repr_id += 1;
@@ -455,6 +455,8 @@  static int ice_eswitch_enable_switchdev(struct ice_pf *pf)
 		return -ENODEV;
 
 	ctrl_vsi = pf->eswitch.control_vsi;
+	/* cp VSI is createad with 1 queue as default */
+	pf->eswitch.qs.value = 1;
 	pf->eswitch.uplink_vsi = uplink_vsi;
 
 	if (ice_eswitch_setup_env(pf))
@@ -487,6 +489,7 @@  static void ice_eswitch_disable_switchdev(struct ice_pf *pf)
 	ice_vsi_release(ctrl_vsi);
 
 	pf->eswitch.is_running = false;
+	pf->eswitch.qs.is_reaching = false;
 }
 
 /**
@@ -615,15 +618,33 @@  static void
 ice_eswitch_cp_change_queues(struct ice_eswitch *eswitch, int change)
 {
 	struct ice_vsi *cp = eswitch->control_vsi;
+	int queues = 0;
+
+	if (eswitch->qs.is_reaching) {
+		if (eswitch->qs.to_reach >= eswitch->qs.value + change) {
+			queues = eswitch->qs.to_reach;
+			eswitch->qs.is_reaching = false;
+		} else {
+			queues = 0;
+		}
+	} else if ((change > 0 && cp->alloc_txq <= eswitch->qs.value) ||
+		   change < 0) {
+		queues = cp->alloc_txq + change;
+	}
 
-	ice_vsi_close(cp);
+	if (queues) {
+		cp->req_txq = queues;
+		cp->req_rxq = queues;
+		ice_vsi_close(cp);
+		ice_vsi_rebuild(cp, ICE_VSI_FLAG_NO_INIT);
+		ice_vsi_open(cp);
+	} else if (!change) {
+		/* change == 0 means that VSI wasn't open, open it here */
+		ice_vsi_open(cp);
+	}
 
-	cp->req_txq = cp->alloc_txq + change;
-	cp->req_rxq = cp->alloc_rxq + change;
-	ice_vsi_rebuild(cp, ICE_VSI_FLAG_NO_INIT);
+	eswitch->qs.value += change;
 	ice_eswitch_remap_rings_to_vectors(eswitch);
-
-	ice_vsi_open(cp);
 }
 
 int
@@ -641,6 +662,7 @@  ice_eswitch_attach(struct ice_pf *pf, struct ice_vf *vf)
 		if (err)
 			return err;
 		/* Control plane VSI is created with 1 queue as default */
+		pf->eswitch.qs.to_reach -= 1;
 		change = 0;
 	}
 
@@ -732,3 +754,19 @@  int ice_eswitch_rebuild(struct ice_pf *pf)
 
 	return 0;
 }
+
+/**
+ * ice_eswitch_reserve_cp_queues - reserve control plane VSI queues
+ * @pf: pointer to PF structure
+ * @change: how many more (or less) queues is needed
+ *
+ * Remember to call ice_eswitch_attach/detach() the "change" times.
+ */
+void ice_eswitch_reserve_cp_queues(struct ice_pf *pf, int change)
+{
+	if (pf->eswitch.qs.value + change < 0)
+		return;
+
+	pf->eswitch.qs.to_reach = pf->eswitch.qs.value + change;
+	pf->eswitch.qs.is_reaching = true;
+}
diff --git a/drivers/net/ethernet/intel/ice/ice_eswitch.h b/drivers/net/ethernet/intel/ice/ice_eswitch.h
index 59d51c0d14e5..1a288a03a79a 100644
--- a/drivers/net/ethernet/intel/ice/ice_eswitch.h
+++ b/drivers/net/ethernet/intel/ice/ice_eswitch.h
@@ -26,6 +26,7 @@  void ice_eswitch_set_target_vsi(struct sk_buff *skb,
 				struct ice_tx_offload_params *off);
 netdev_tx_t
 ice_eswitch_port_start_xmit(struct sk_buff *skb, struct net_device *netdev);
+void ice_eswitch_reserve_cp_queues(struct ice_pf *pf, int change);
 #else /* CONFIG_ICE_SWITCHDEV */
 static inline void ice_eswitch_detach(struct ice_pf *pf, struct ice_vf *vf) { }
 
@@ -76,5 +77,8 @@  ice_eswitch_port_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 {
 	return NETDEV_TX_BUSY;
 }
+
+static inline void
+ice_eswitch_reserve_cp_queues(struct ice_pf *pf, int change) { }
 #endif /* CONFIG_ICE_SWITCHDEV */
 #endif /* _ICE_ESWITCH_H_ */
diff --git a/drivers/net/ethernet/intel/ice/ice_sriov.c b/drivers/net/ethernet/intel/ice/ice_sriov.c
index 51f5f420d632..5a45bd5ce6ad 100644
--- a/drivers/net/ethernet/intel/ice/ice_sriov.c
+++ b/drivers/net/ethernet/intel/ice/ice_sriov.c
@@ -172,6 +172,8 @@  void ice_free_vfs(struct ice_pf *pf)
 	else
 		dev_warn(dev, "VFs are assigned - not disabling SR-IOV\n");
 
+	ice_eswitch_reserve_cp_queues(pf, -ice_get_num_vfs(pf));
+
 	mutex_lock(&vfs->table_lock);
 
 	ice_for_each_vf(pf, bkt, vf) {
@@ -930,6 +932,7 @@  static int ice_ena_vfs(struct ice_pf *pf, u16 num_vfs)
 		goto err_unroll_sriov;
 	}
 
+	ice_eswitch_reserve_cp_queues(pf, num_vfs);
 	ret = ice_start_vfs(pf);
 	if (ret) {
 		dev_err(dev, "Failed to start %d VFs, err %d\n", num_vfs, ret);