Merge tag 'mlx5-fixes-2022-11-21' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5 fixes 2022-11-21

This series provides bug fixes to mlx5 driver.

* tag 'mlx5-fixes-2022-11-21' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
net/mlx5e: Fix possible race condition in macsec extended packet number update routine
net/mlx5e: Fix MACsec update SecY
net/mlx5e: Fix MACsec SA initialization routine
net/mlx5e: Remove leftovers from old XSK queues enumeration
net/mlx5e: Offload rule only when all encaps are valid
net/mlx5e: Fix missing alignment in size of MTT/KLM entries
net/mlx5: Fix sync reset event handler error flow
net/mlx5: E-Switch, Set correctly vport destination
net/mlx5: Lag, avoid lockdep warnings
net/mlx5: Fix handling of entry refcount when command is not issued to FW
net/mlx5: cmdif, Print info on any firmware cmd failure to tracepoint
net/mlx5: SF: Fix probing active SFs during driver probe phase
net/mlx5: Fix FW tracer timestamp calculation
net/mlx5: Do not query pci info while pci disabled
====================

Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

drivers/net/ethernet/mellanox/mlx5/core/cmd.c

···
 #include "mlx5_core.h"
 #include "lib/eq.h"
 #include "lib/tout.h"
+#define CREATE_TRACE_POINTS
+#include "diag/cmd_tracepoint.h"
 
 enum {
 	CMD_IF_REV = 5,
···
 static void cmd_status_print(struct mlx5_core_dev *dev, void *in, void *out)
 {
 	u16 opcode, op_mod;
-	u32 syndrome;
-	u8  status;
 	u16 uid;
-	int err;
-
-	syndrome = MLX5_GET(mbox_out, out, syndrome);
-	status = MLX5_GET(mbox_out, out, status);
 
 	opcode = MLX5_GET(mbox_in, in, opcode);
 	op_mod = MLX5_GET(mbox_in, in, op_mod);
 	uid    = MLX5_GET(mbox_in, in, uid);
 
-	err = cmd_status_to_err(status);
-
 	if (!uid && opcode != MLX5_CMD_OP_DESTROY_MKEY)
 		mlx5_cmd_out_err(dev, opcode, op_mod, out);
-	else
-		mlx5_core_dbg(dev,
-			"%s(0x%x) op_mod(0x%x) uid(%d) failed, status %s(0x%x), syndrome (0x%x), err(%d)\n",
-			mlx5_command_str(opcode), opcode, op_mod, uid,
-			cmd_status_str(status), status, syndrome, err);
 }
 
 int mlx5_cmd_check(struct mlx5_core_dev *dev, int err, void *in, void *out)
···
 		cmd_ent_get(ent);
 	set_bit(MLX5_CMD_ENT_STATE_PENDING_COMP, &ent->state);
 
+	cmd_ent_get(ent); /* for the _real_ FW event on completion */
 	/* Skip sending command to fw if internal error */
 	if (mlx5_cmd_is_down(dev) || !opcode_allowed(&dev->cmd, ent->op)) {
 		ent->ret = -ENXIO;
···
 		return;
 	}
 
-	cmd_ent_get(ent); /* for the _real_ FW event on completion */
 	/* ring doorbell after the descriptor is valid */
 	mlx5_core_dbg(dev, "writing 0x%x to command doorbell\n", 1 << ent->idx);
 	wmb();
···
 				cmd_ent_put(ent); /* timeout work was canceled */
 
 			if (!forced || /* Real FW completion */
-			    pci_channel_offline(dev->pdev) || /* FW is inaccessible */
-			    dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
+			     mlx5_cmd_is_down(dev) || /* No real FW completion is expected */
+			     !opcode_allowed(cmd, ent->op))
 				cmd_ent_put(ent);
 
 			ent->ts2 = ktime_get_ns();
···
 	return err;
 }
 
+static void mlx5_cmd_err_trace(struct mlx5_core_dev *dev, u16 opcode, u16 op_mod, void *out)
+{
+	u32 syndrome = MLX5_GET(mbox_out, out, syndrome);
+	u8 status = MLX5_GET(mbox_out, out, status);
+
+	trace_mlx5_cmd(mlx5_command_str(opcode), opcode, op_mod,
+		       cmd_status_str(status), status, syndrome,
+		       cmd_status_to_err(status));
+}
+
 static void cmd_status_log(struct mlx5_core_dev *dev, u16 opcode, u8 status,
 			   u32 syndrome, int err)
 {
···
 }
 
 /* preserve -EREMOTEIO for outbox.status != OK, otherwise return err as is */
-static int cmd_status_err(struct mlx5_core_dev *dev, int err, u16 opcode, void *out)
+static int cmd_status_err(struct mlx5_core_dev *dev, int err, u16 opcode, u16 op_mod, void *out)
 {
 	u32 syndrome = MLX5_GET(mbox_out, out, syndrome);
 	u8 status = MLX5_GET(mbox_out, out, status);
···
 	if (err == -EREMOTEIO) /* -EREMOTEIO is preserved */
 		err = -EIO;
 
-	if (!err && status != MLX5_CMD_STAT_OK)
+	if (!err && status != MLX5_CMD_STAT_OK) {
 		err = -EREMOTEIO;
+		mlx5_cmd_err_trace(dev, opcode, op_mod, out);
+	}
 
 	cmd_status_log(dev, opcode, status, syndrome, err);
 	return err;
···
 {
 	int err = cmd_exec(dev, in, in_size, out, out_size, NULL, NULL, false);
 	u16 opcode = MLX5_GET(mbox_in, in, opcode);
+	u16 op_mod = MLX5_GET(mbox_in, in, op_mod);
 
-	err = cmd_status_err(dev, err, opcode, out);
-	return err;
+	return cmd_status_err(dev, err, opcode, op_mod, out);
 }
 EXPORT_SYMBOL(mlx5_cmd_do);
 
···
 {
 	int err = cmd_exec(dev, in, in_size, out, out_size, NULL, NULL, true);
 	u16 opcode = MLX5_GET(mbox_in, in, opcode);
+	u16 op_mod = MLX5_GET(mbox_in, in, op_mod);
 
-	err = cmd_status_err(dev, err, opcode, out);
+	err = cmd_status_err(dev, err, opcode, op_mod, out);
 	return mlx5_cmd_check(dev, err, in, out);
 }
 EXPORT_SYMBOL(mlx5_cmd_exec_polling);
···
 	struct mlx5_async_ctx *ctx;
 
 	ctx = work->ctx;
-	status = cmd_status_err(ctx->dev, status, work->opcode, work->out);
+	status = cmd_status_err(ctx->dev, status, work->opcode, work->op_mod, work->out);
 	work->user_callback(status, work);
 	if (atomic_dec_and_test(&ctx->num_inflight))
 		complete(&ctx->inflight_done);
···
 	work->ctx = ctx;
 	work->user_callback = callback;
 	work->opcode = MLX5_GET(mbox_in, in, opcode);
+	work->op_mod = MLX5_GET(mbox_in, in, op_mod);
 	work->out = out;
 	if (WARN_ON(!atomic_inc_not_zero(&ctx->num_inflight)))
 		return -EIO;

drivers/net/ethernet/mellanox/mlx5/core/diag/cmd_tracepoint.h

···
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved. */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM mlx5
+
+#if !defined(_MLX5_CMD_TP_H_) || defined(TRACE_HEADER_MULTI_READ)
+#define _MLX5_CMD_TP_H_
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+TRACE_EVENT(mlx5_cmd,
+	    TP_PROTO(const char *command_str, u16 opcode, u16 op_mod,
+		     const char *status_str, u8 status, u32 syndrome, int err),
+	    TP_ARGS(command_str, opcode, op_mod, status_str, status, syndrome, err),
+	    TP_STRUCT__entry(__string(command_str, command_str)
+			     __field(u16, opcode)
+			     __field(u16, op_mod)
+			    __string(status_str, status_str)
+			    __field(u8, status)
+			    __field(u32, syndrome)
+			    __field(int, err)
+			    ),
+	    TP_fast_assign(__assign_str(command_str, command_str);
+			__entry->opcode = opcode;
+			__entry->op_mod = op_mod;
+			__assign_str(status_str, status_str);
+			__entry->status = status;
+			__entry->syndrome = syndrome;
+			__entry->err = err;
+	    ),
+	    TP_printk("%s(0x%x) op_mod(0x%x) failed, status %s(0x%x), syndrome (0x%x), err(%d)",
+		      __get_str(command_str), __entry->opcode, __entry->op_mod,
+		      __get_str(status_str), __entry->status, __entry->syndrome,
+		      __entry->err)
+);
+
+#endif /* _MLX5_CMD_TP_H_ */
+
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH ./diag
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE cmd_tracepoint
+#include <trace/define_trace.h>

drivers/net/ethernet/mellanox/mlx5/core/diag/fw_tracer.c

···
 			trace_timestamp = (timestamp_event.timestamp & MASK_52_7) |
 					  (str_frmt->timestamp & MASK_6_0);
 		else
-			trace_timestamp = ((timestamp_event.timestamp & MASK_52_7) - 1) |
+			trace_timestamp = ((timestamp_event.timestamp - 1) & MASK_52_7) |
 					  (str_frmt->timestamp & MASK_6_0);
 
 		mlx5_tracer_print_trace(str_frmt, dev, trace_timestamp);

drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_encap.c

···
 	list_for_each_entry(flow, flow_list, tmp_list) {
 		if (!mlx5e_is_offloaded_flow(flow) || flow_flag_test(flow, SLOW))
 			continue;
-		spec = &flow->attr->parse_attr->spec;
-
-		/* update from encap rule to slow path rule */
-		rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec);
 
 		attr = mlx5e_tc_get_encap_attr(flow);
 		esw_attr = attr->esw_attr;
 		/* mark the flow's encap dest as non-valid */
 		esw_attr->dests[flow->tmp_entry_index].flags &= ~MLX5_ESW_DEST_ENCAP_VALID;
+		esw_attr->dests[flow->tmp_entry_index].pkt_reformat = NULL;
+
+		/* update from encap rule to slow path rule */
+		spec = &flow->attr->parse_attr->spec;
+		rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec);
 
 		if (IS_ERR(rule)) {
 			err = PTR_ERR(rule);
···
 	/* we know that the encap is valid */
 	e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
 	mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
+	e->pkt_reformat = NULL;
 }
 
 static void mlx5e_take_tmp_flow(struct mlx5e_tc_flow *flow,
···
 		       struct net_device *mirred_dev,
 		       int out_index,
 		       struct netlink_ext_ack *extack,
-		       struct net_device **encap_dev,
-		       bool *encap_valid)
+		       struct net_device **encap_dev)
 {
 	struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 	struct mlx5e_tc_flow_parse_attr *parse_attr;
···
 	if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
 		attr->esw_attr->dests[out_index].pkt_reformat = e->pkt_reformat;
 		attr->esw_attr->dests[out_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
-		*encap_valid = true;
 	} else {
-		*encap_valid = false;
+		flow_flag_set(flow, SLOW);
 	}
 	mutex_unlock(&esw->offloads.encap_tbl_lock);
 

drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_encap.h

···
 		       struct net_device *mirred_dev,
 		       int out_index,
 		       struct netlink_ext_ack *extack,
-		       struct net_device **encap_dev,
-		       bool *encap_valid);
+		       struct net_device **encap_dev);
 
 int mlx5e_attach_decap(struct mlx5e_priv *priv,
 		       struct mlx5e_tc_flow *flow,

drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c

···
 	obj_attrs.aso_pdn = macsec->aso.pdn;
 	obj_attrs.epn_state = sa->epn_state;
 
-	if (is_tx) {
-		obj_attrs.ssci = cpu_to_be32((__force u32)ctx->sa.tx_sa->ssci);
-		key = &ctx->sa.tx_sa->key;
-	} else {
-		obj_attrs.ssci = cpu_to_be32((__force u32)ctx->sa.rx_sa->ssci);
-		key = &ctx->sa.rx_sa->key;
+	key = (is_tx) ? &ctx->sa.tx_sa->key : &ctx->sa.rx_sa->key;
+
+	if (sa->epn_state.epn_enabled) {
+		obj_attrs.ssci = (is_tx) ? cpu_to_be32((__force u32)ctx->sa.tx_sa->ssci) :
+					   cpu_to_be32((__force u32)ctx->sa.rx_sa->ssci);
+
+		memcpy(&obj_attrs.salt, &key->salt, sizeof(key->salt));
 	}
 
-	memcpy(&obj_attrs.salt, &key->salt, sizeof(key->salt));
 	obj_attrs.replay_window = ctx->secy->replay_window;
 	obj_attrs.replay_protect = ctx->secy->replay_protect;
 
···
 				continue;
 
 			if (rx_sa->active) {
-				err = mlx5e_macsec_init_sa(ctx, rx_sa, false, false);
+				err = mlx5e_macsec_init_sa(ctx, rx_sa, true, false);
 				if (err)
 					goto out;
 			}
···
 
 	async_work = container_of(work, struct mlx5e_macsec_async_work, work);
 	macsec = async_work->macsec;
+	mutex_lock(&macsec->lock);
+
 	mdev = async_work->mdev;
 	obj_id = async_work->obj_id;
 	macsec_sa = get_macsec_tx_sa_from_obj_id(macsec, obj_id);
···
 
 out_async_work:
 	kfree(async_work);
+	mutex_unlock(&macsec->lock);
 }
 
 static int macsec_obj_change_event(struct notifier_block *nb, unsigned long event, void *data)

drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c

···
 #include "en.h"
 #include "en/port.h"
 #include "en/params.h"
-#include "en/xsk/pool.h"
 #include "en/ptp.h"
 #include "lib/clock.h"
 #include "en/fs_ethtool.h"
···
 				struct ethtool_channels *ch)
 {
 	mutex_lock(&priv->state_lock);
-
 	ch->max_combined   = priv->max_nch;
 	ch->combined_count = priv->channels.params.num_channels;
-	if (priv->xsk.refcnt) {
-		/* The upper half are XSK queues. */
-		ch->max_combined *= 2;
-		ch->combined_count *= 2;
-	}
-
 	mutex_unlock(&priv->state_lock);
 }
 
···
 		return 0;
 
 	mutex_lock(&priv->state_lock);
-
-	/* Don't allow changing the number of channels if there is an active
-	 * XSK, because the numeration of the XSK and regular RQs will change.
-	 */
-	if (priv->xsk.refcnt) {
-		err = -EINVAL;
-		netdev_err(priv->netdev, "%s: AF_XDP is active, cannot change the number of channels\n",
-			   __func__);
-		goto out;
-	}
 
 	/* Don't allow changing the number of channels if HTB offload is active,
 	 * because the numeration of the QoS SQs will change, while per-queue

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

···
 static u16 mlx5e_mpwrq_umr_octowords(u32 entries, enum mlx5e_mpwrq_umr_mode umr_mode)
 {
 	u8 umr_entry_size = mlx5e_mpwrq_umr_entry_size(umr_mode);
+	u32 sz;
 
-	WARN_ON_ONCE(entries * umr_entry_size % MLX5_OCTWORD);
+	sz = ALIGN(entries * umr_entry_size, MLX5_UMR_MTT_ALIGNMENT);
 
-	return entries * umr_entry_size / MLX5_OCTWORD;
+	return sz / MLX5_OCTWORD;
 }
 
 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

···
 		struct mlx5e_tc_flow *flow,
 		struct mlx5_flow_attr *attr,
 		struct netlink_ext_ack *extack,
-		bool *encap_valid,
 		bool *vf_tun)
 {
 	struct mlx5e_tc_flow_parse_attr *parse_attr;
···
 	parse_attr = attr->parse_attr;
 	esw_attr = attr->esw_attr;
 	*vf_tun = false;
-	*encap_valid = true;
 
 	for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
 		struct net_device *out_dev;
···
 			goto out;
 		}
 		err = mlx5e_attach_encap(priv, flow, attr, out_dev, out_index,
-					 extack, &encap_dev, encap_valid);
+					 extack, &encap_dev);
 		dev_put(out_dev);
 		if (err)
 			goto out;
···
 	struct mlx5e_tc_flow_parse_attr *parse_attr;
 	struct mlx5_flow_attr *attr = flow->attr;
 	struct mlx5_esw_flow_attr *esw_attr;
-	bool vf_tun, encap_valid;
 	u32 max_prio, max_chain;
+	bool vf_tun;
 	int err = 0;
 
 	parse_attr = attr->parse_attr;
···
 		esw_attr->int_port = int_port;
 	}
 
-	err = set_encap_dests(priv, flow, attr, extack, &encap_valid, &vf_tun);
+	err = set_encap_dests(priv, flow, attr, extack, &vf_tun);
 	if (err)
 		goto err_out;
 
···
 	 * (1) there's no error
 	 * (2) there's an encap action and we don't have valid neigh
 	 */
-	if (!encap_valid || flow_flag_test(flow, SLOW))
+	if (flow_flag_test(flow, SLOW))
 		flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec);
 	else
 		flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
···
 	struct mlx5e_post_act *post_act = get_post_action(flow->priv);
 	struct mlx5_flow_attr *attr, *next_attr = NULL;
 	struct mlx5e_post_act_handle *handle;
-	bool vf_tun, encap_valid = true;
+	bool vf_tun;
 	int err;
 
 	/* This is going in reverse order as needed.
···
 		if (list_is_last(&attr->list, &flow->attrs))
 			break;
 
-		err = set_encap_dests(flow->priv, flow, attr, extack, &encap_valid, &vf_tun);
+		err = set_encap_dests(flow->priv, flow, attr, extack, &vf_tun);
 		if (err)
 			goto out_free;
-
-		if (!encap_valid)
-			flow_flag_set(flow, SLOW);
 
 		err = actions_prepare_mod_hdr_actions(flow->priv, flow, attr, extack);
 		if (err)

drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c

···
 		    mlx5_lag_mpesw_is_activated(esw->dev))
 			dest[dest_idx].type = MLX5_FLOW_DESTINATION_TYPE_UPLINK;
 	}
-	if (esw_attr->dests[attr_idx].flags & MLX5_ESW_DEST_ENCAP) {
+	if (esw_attr->dests[attr_idx].flags & MLX5_ESW_DEST_ENCAP_VALID) {
 		if (pkt_reformat) {
 			flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
 			flow_act->pkt_reformat = esw_attr->dests[attr_idx].pkt_reformat;

drivers/net/ethernet/mellanox/mlx5/core/fw_reset.c

···
 	MLX5_FW_RESET_FLAGS_RESET_REQUESTED,
 	MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST,
 	MLX5_FW_RESET_FLAGS_PENDING_COMP,
-	MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS
+	MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS,
+	MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED
 };
 
 struct mlx5_fw_reset {
···
 	err = mlx5_pci_link_toggle(dev);
 	if (err) {
 		mlx5_core_warn(dev, "mlx5_pci_link_toggle failed, no reset done, err %d\n", err);
-		goto done;
+		set_bit(MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED, &fw_reset->reset_flags);
 	}
 
 	mlx5_enter_error_state(dev, true);
···
 		goto out;
 	}
 	err = fw_reset->ret;
+	if (test_and_clear_bit(MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED, &fw_reset->reset_flags)) {
+		mlx5_unload_one_devl_locked(dev);
+		mlx5_load_one_devl_locked(dev, false);
+	}
 out:
 	clear_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
 	return err;

drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c

···
 	if (ldev->nb.notifier_call)
 		unregister_netdevice_notifier_net(&init_net, &ldev->nb);
 	mlx5_lag_mp_cleanup(ldev);
-	mlx5_lag_mpesw_cleanup(ldev);
-	cancel_work_sync(&ldev->mpesw_work);
 	destroy_workqueue(ldev->wq);
+	mlx5_lag_mpesw_cleanup(ldev);
 	mutex_destroy(&ldev->lock);
 	kfree(ldev);
 }

drivers/net/ethernet/mellanox/mlx5/core/lag/lag.h

···
 	enum netdev_lag_hash hash_type;
 };
 
+enum mpesw_op {
+	MLX5_MPESW_OP_ENABLE,
+	MLX5_MPESW_OP_DISABLE,
+};
+
+struct mlx5_mpesw_work_st {
+	struct work_struct work;
+	struct mlx5_lag    *lag;
+	enum mpesw_op	   op;
+	struct completion  comp;
+	int result;
+};
+
 /* LAG data of a ConnectX card.
  * It serves both its phys functions.
  */
···
 	struct lag_tracker        tracker;
 	struct workqueue_struct   *wq;
 	struct delayed_work       bond_work;
-	struct work_struct	  mpesw_work;
 	struct notifier_block     nb;
 	struct lag_mp             lag_mp;
 	struct mlx5_lag_port_sel  port_sel;

drivers/net/ethernet/mellanox/mlx5/core/lag/mpesw.c

···
 #include "eswitch.h"
 #include "lib/mlx5.h"
 
-void mlx5_mpesw_work(struct work_struct *work)
+static int add_mpesw_rule(struct mlx5_lag *ldev)
 {
-	struct mlx5_lag *ldev = container_of(work, struct mlx5_lag, mpesw_work);
+	struct mlx5_core_dev *dev = ldev->pf[MLX5_LAG_P1].dev;
+	int err;
 
-	mutex_lock(&ldev->lock);
-	mlx5_disable_lag(ldev);
-	mutex_unlock(&ldev->lock);
+	if (atomic_add_return(1, &ldev->lag_mpesw.mpesw_rule_count) != 1)
+		return 0;
+
+	if (ldev->mode != MLX5_LAG_MODE_NONE) {
+		err = -EINVAL;
+		goto out_err;
+	}
+
+	err = mlx5_activate_lag(ldev, NULL, MLX5_LAG_MODE_MPESW, false);
+	if (err) {
+		mlx5_core_warn(dev, "Failed to create LAG in MPESW mode (%d)\n", err);
+		goto out_err;
+	}
+
+	return 0;
+
+out_err:
+	atomic_dec(&ldev->lag_mpesw.mpesw_rule_count);
+	return err;
 }
 
-static void mlx5_lag_disable_mpesw(struct mlx5_core_dev *dev)
+static void del_mpesw_rule(struct mlx5_lag *ldev)
 {
-	struct mlx5_lag *ldev = dev->priv.lag;
-
-	if (!queue_work(ldev->wq, &ldev->mpesw_work))
-		mlx5_core_warn(dev, "failed to queue work\n");
-}
-
-void mlx5_lag_del_mpesw_rule(struct mlx5_core_dev *dev)
-{
-	struct mlx5_lag *ldev = dev->priv.lag;
-
-	if (!ldev)
-		return;
-
-	mutex_lock(&ldev->lock);
 	if (!atomic_dec_return(&ldev->lag_mpesw.mpesw_rule_count) &&
 	    ldev->mode == MLX5_LAG_MODE_MPESW)
-		mlx5_lag_disable_mpesw(dev);
-	mutex_unlock(&ldev->lock);
+		mlx5_disable_lag(ldev);
 }
 
-int mlx5_lag_add_mpesw_rule(struct mlx5_core_dev *dev)
+static void mlx5_mpesw_work(struct work_struct *work)
+{
+	struct mlx5_mpesw_work_st *mpesww = container_of(work, struct mlx5_mpesw_work_st, work);
+	struct mlx5_lag *ldev = mpesww->lag;
+
+	mutex_lock(&ldev->lock);
+	if (mpesww->op == MLX5_MPESW_OP_ENABLE)
+		mpesww->result = add_mpesw_rule(ldev);
+	else if (mpesww->op == MLX5_MPESW_OP_DISABLE)
+		del_mpesw_rule(ldev);
+	mutex_unlock(&ldev->lock);
+
+	complete(&mpesww->comp);
+}
+
+static int mlx5_lag_mpesw_queue_work(struct mlx5_core_dev *dev,
+				     enum mpesw_op op)
 {
 	struct mlx5_lag *ldev = dev->priv.lag;
+	struct mlx5_mpesw_work_st *work;
 	int err = 0;
 
 	if (!ldev)
 		return 0;
 
-	mutex_lock(&ldev->lock);
-	if (atomic_add_return(1, &ldev->lag_mpesw.mpesw_rule_count) != 1)
-		goto out;
+	work = kzalloc(sizeof(*work), GFP_KERNEL);
+	if (!work)
+		return -ENOMEM;
 
-	if (ldev->mode != MLX5_LAG_MODE_NONE) {
+	INIT_WORK(&work->work, mlx5_mpesw_work);
+	init_completion(&work->comp);
+	work->op = op;
+	work->lag = ldev;
+
+	if (!queue_work(ldev->wq, &work->work)) {
+		mlx5_core_warn(dev, "failed to queue mpesw work\n");
 		err = -EINVAL;
 		goto out;
 	}
-
-	err = mlx5_activate_lag(ldev, NULL, MLX5_LAG_MODE_MPESW, false);
-	if (err)
-		mlx5_core_warn(dev, "Failed to create LAG in MPESW mode (%d)\n", err);
-
+	wait_for_completion(&work->comp);
+	err = work->result;
 out:
-	mutex_unlock(&ldev->lock);
+	kfree(work);
 	return err;
+}
+
+void mlx5_lag_del_mpesw_rule(struct mlx5_core_dev *dev)
+{
+	mlx5_lag_mpesw_queue_work(dev, MLX5_MPESW_OP_DISABLE);
+}
+
+int mlx5_lag_add_mpesw_rule(struct mlx5_core_dev *dev)
+{
+	return mlx5_lag_mpesw_queue_work(dev, MLX5_MPESW_OP_ENABLE);
 }
 
 int mlx5_lag_do_mirred(struct mlx5_core_dev *mdev, struct net_device *out_dev)
···
 	if (!netif_is_bond_master(out_dev) || !ldev)
 		return 0;
 
-	mutex_lock(&ldev->lock);
-	if (ldev->mode == MLX5_LAG_MODE_MPESW) {
-		mutex_unlock(&ldev->lock);
+	if (ldev->mode == MLX5_LAG_MODE_MPESW)
 		return -EOPNOTSUPP;
-	}
-	mutex_unlock(&ldev->lock);
+
 	return 0;
 }
 
···
 
 void mlx5_lag_mpesw_init(struct mlx5_lag *ldev)
 {
-	INIT_WORK(&ldev->mpesw_work, mlx5_mpesw_work);
 	atomic_set(&ldev->lag_mpesw.mpesw_rule_count, 0);
 }
 
 void mlx5_lag_mpesw_cleanup(struct mlx5_lag *ldev)
 {
-	cancel_delayed_work_sync(&ldev->bond_work);
+	WARN_ON(atomic_read(&ldev->lag_mpesw.mpesw_rule_count));
 }

drivers/net/ethernet/mellanox/mlx5/core/lag/mpesw.h

···
 	atomic_t mpesw_rule_count;
 };
 
-void mlx5_mpesw_work(struct work_struct *work);
 int mlx5_lag_do_mirred(struct mlx5_core_dev *mdev, struct net_device *out_dev);
 bool mlx5_lag_mpesw_is_activated(struct mlx5_core_dev *dev);
 #if IS_ENABLED(CONFIG_MLX5_ESWITCH)

drivers/net/ethernet/mellanox/mlx5/core/main.c

···
 	res = state == pci_channel_io_perm_failure ?
 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
 
-	mlx5_pci_trace(dev, "Exit, result = %d, %s\n",  res, result2str(res));
+	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
+		       __func__, dev->state, dev->pci_status, res, result2str(res));
 	return res;
 }
 
···
 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
 	int err;
 
-	mlx5_pci_trace(dev, "Enter\n");
+	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
+		       __func__, dev->state, dev->pci_status);
 
 	err = mlx5_pci_enable_device(dev);
 	if (err) {
···
 
 	res = PCI_ERS_RESULT_RECOVERED;
 out:
-	mlx5_pci_trace(dev, "Exit, err = %d, result = %d, %s\n", err, res, result2str(res));
+	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
+		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
 	return res;
 }
 

drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c

···
 	phys_addr_t base_address;
 	u64 sf_bar_length;
 	struct notifier_block nb;
+	struct mutex table_lock; /* Serializes sf life cycle and vhca state change handler */
+	struct workqueue_struct *active_wq;
+	struct work_struct work;
+	u8 stop_active_wq:1;
 	struct mlx5_core_dev *dev;
 };
 
···
 		return 0;
 
 	sf_index = event->function_id - base_id;
+	mutex_lock(&table->table_lock);
 	sf_dev = xa_load(&table->devices, sf_index);
 	switch (event->new_vhca_state) {
 	case MLX5_VHCA_STATE_INVALID:
···
 	default:
 		break;
 	}
+	mutex_unlock(&table->table_lock);
 	return 0;
 }
 
···
 		function_id++;
 	}
 	return 0;
+}
+
+static void mlx5_sf_dev_add_active_work(struct work_struct *work)
+{
+	struct mlx5_sf_dev_table *table = container_of(work, struct mlx5_sf_dev_table, work);
+	u32 out[MLX5_ST_SZ_DW(query_vhca_state_out)] = {};
+	struct mlx5_core_dev *dev = table->dev;
+	u16 max_functions;
+	u16 function_id;
+	u16 sw_func_id;
+	int err = 0;
+	u8 state;
+	int i;
+
+	max_functions = mlx5_sf_max_functions(dev);
+	function_id = MLX5_CAP_GEN(dev, sf_base_id);
+	for (i = 0; i < max_functions; i++, function_id++) {
+		if (table->stop_active_wq)
+			return;
+		err = mlx5_cmd_query_vhca_state(dev, function_id, out, sizeof(out));
+		if (err)
+			/* A failure of specific vhca doesn't mean others will
+			 * fail as well.
+			 */
+			continue;
+		state = MLX5_GET(query_vhca_state_out, out, vhca_state_context.vhca_state);
+		if (state != MLX5_VHCA_STATE_ACTIVE)
+			continue;
+
+		sw_func_id = MLX5_GET(query_vhca_state_out, out, vhca_state_context.sw_function_id);
+		mutex_lock(&table->table_lock);
+		/* Don't probe device which is already probe */
+		if (!xa_load(&table->devices, i))
+			mlx5_sf_dev_add(dev, i, function_id, sw_func_id);
+		/* There is a race where SF got inactive after the query
+		 * above. e.g.: the query returns that the state of the
+		 * SF is active, and after that the eswitch manager set it to
+		 * inactive.
+		 * This case cannot be managed in SW, since the probing of the
+		 * SF is on one system, and the inactivation is on a different
+		 * system.
+		 * If the inactive is done after the SF perform init_hca(),
+		 * the SF will fully probe and then removed. If it was
+		 * done before init_hca(), the SF probe will fail.
+		 */
+		mutex_unlock(&table->table_lock);
+	}
+}
+
+/* In case SFs are generated externally, probe active SFs */
+static int mlx5_sf_dev_queue_active_work(struct mlx5_sf_dev_table *table)
+{
+	if (MLX5_CAP_GEN(table->dev, eswitch_manager))
+		return 0; /* the table is local */
+
+	/* Use a workqueue to probe active SFs, which are in large
+	 * quantity and may take up to minutes to probe.
+	 */
+	table->active_wq = create_singlethread_workqueue("mlx5_active_sf");
+	if (!table->active_wq)
+		return -ENOMEM;
+	INIT_WORK(&table->work, &mlx5_sf_dev_add_active_work);
+	queue_work(table->active_wq, &table->work);
+	return 0;
+}
+
+static void mlx5_sf_dev_destroy_active_work(struct mlx5_sf_dev_table *table)
+{
+	if (table->active_wq) {
+		table->stop_active_wq = true;
+		destroy_workqueue(table->active_wq);
+	}
 }
 
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
···
 	table->base_address = pci_resource_start(dev->pdev, 2);
 	table->max_sfs = max_sfs;
 	xa_init(&table->devices);
+	mutex_init(&table->table_lock);
 	dev->priv.sf_dev_table = table;
 
 	err = mlx5_vhca_event_notifier_register(dev, &table->nb);
 	if (err)
 		goto vhca_err;
+
+	err = mlx5_sf_dev_queue_active_work(table);
+	if (err)
+		goto add_active_err;
+
 	err = mlx5_sf_dev_vhca_arm_all(table);
 	if (err)
 		goto arm_err;
···
 	return;
 
 arm_err:
+	mlx5_sf_dev_destroy_active_work(table);
+add_active_err:
 	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
 vhca_err:
 	table->max_sfs = 0;
···
 	if (!table)
 		return;
 
+	mlx5_sf_dev_destroy_active_work(table);
 	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
+	mutex_destroy(&table->table_lock);
 
 	/* Now that event handler is not running, it is safe to destroy
 	 * the sf device without race.

include/linux/mlx5/driver.h

···
 	struct mlx5_async_ctx *ctx;
 	mlx5_async_cbk_t user_callback;
 	u16 opcode; /* cmd opcode */
+	u16 op_mod; /* cmd op_mod */
 	void *out; /* pointer to the cmd output buffer */
 };