Merge branch 'mlx5-fixes'

Saeed Mahameed says:

====================
Mellanox 100G mlx5 fixes 2016-10-25

This series contains some bug fixes for the mlx5 core and mlx5e driver.

From Daniel:
    - Cache line size determination at runtime, instead of using
      L1_CACHE_BYTES hard coded value, use cache_line_size()
    - Always Query HCA caps after setting them even on reset flow

From Mohamad:
    - Reorder netdev cleanup to uregister netdev before detaching it
      for the kernel to not complain about open resources such as vlans
    - Change the acl enable prototype to return status, for better error
      resiliency
    - Clear health sick bit when starting health poll after reset flow
    - Fix race between PCI error handlers and health work
    - PCI error recovery health care simulation, in case when the kernel
      PCI error handlers are not triggered for some internal firmware errors

From Noa:
    - Avoid passing dma address 0 to firmware when mapping system pages
      to the firmware

From Paul: Some straight forward flow steering fixes
    - Keep autogroups list ordered
    - Fix autogroups groups num not decreasing
    - Correctly initialize last use of flow counters

From Saeed:
    - Choose the nearest LRO timeout to the wanted one
      instead of blindly choosing "dev_cap.lro_timeout[2]"

This series has no conflict with the for-next pull request posted
earlier today ("Mellanox mlx5 core driver updates 2016-10-25").
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/infiniband/hw/mlx5/main.c

@@ -1019,7 +1019,7 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
 	resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);
 	if (mlx5_core_is_pf(dev->mdev) && MLX5_CAP_GEN(dev->mdev, bf))
 		resp.bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size);
-	resp.cache_line_size = L1_CACHE_BYTES;
+	resp.cache_line_size = cache_line_size();
 	resp.max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq);
 	resp.max_rq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq);
 	resp.max_send_wqebb = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);

drivers/infiniband/hw/mlx5/qp.c

@@ -52,7 +52,6 @@ enum {
 
 enum {
 	MLX5_IB_SQ_STRIDE	= 6,
-	MLX5_IB_CACHE_LINE_SIZE	= 64,
 };
 
 static const u32 mlx5_ib_opcode[] = {

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

@@ -41,6 +41,13 @@
 
 #include "mlx5_core.h"
 
+struct mlx5_db_pgdir {
+	struct list_head	list;
+	unsigned long	       *bitmap;
+	__be32		       *db_page;
+	dma_addr_t		db_dma;
+};
+
 /* Handling for queue buffers -- we allocate a bunch of memory and
  * register it in a memory region at HCA virtual address 0.
  */
@@ -102,17 +109,28 @@ EXPORT_SYMBOL_GPL(mlx5_buf_free);
 static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
 						 int node)
 {
+	u32 db_per_page = PAGE_SIZE / cache_line_size();
 	struct mlx5_db_pgdir *pgdir;
 
 	pgdir = kzalloc(sizeof(*pgdir), GFP_KERNEL);
 	if (!pgdir)
 		return NULL;
 
-	bitmap_fill(pgdir->bitmap, MLX5_DB_PER_PAGE);
+	pgdir->bitmap = kcalloc(BITS_TO_LONGS(db_per_page),
+				sizeof(unsigned long),
+				GFP_KERNEL);
+
+	if (!pgdir->bitmap) {
+		kfree(pgdir);
+		return NULL;
+	}
+
+	bitmap_fill(pgdir->bitmap, db_per_page);
 
 	pgdir->db_page = mlx5_dma_zalloc_coherent_node(dev, PAGE_SIZE,
 						       &pgdir->db_dma, node);
 	if (!pgdir->db_page) {
+		kfree(pgdir->bitmap);
 		kfree(pgdir);
 		return NULL;
 	}
@@ -123,18 +141,19 @@ static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
 static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir,
 				    struct mlx5_db *db)
 {
+	u32 db_per_page = PAGE_SIZE / cache_line_size();
 	int offset;
 	int i;
 
-	i = find_first_bit(pgdir->bitmap, MLX5_DB_PER_PAGE);
-	if (i >= MLX5_DB_PER_PAGE)
+	i = find_first_bit(pgdir->bitmap, db_per_page);
+	if (i >= db_per_page)
 		return -ENOMEM;
 
 	__clear_bit(i, pgdir->bitmap);
 
 	db->u.pgdir = pgdir;
 	db->index   = i;
-	offset = db->index * L1_CACHE_BYTES;
+	offset = db->index * cache_line_size();
 	db->db      = pgdir->db_page + offset / sizeof(*pgdir->db_page);
 	db->dma     = pgdir->db_dma  + offset;
 
@@ -181,14 +200,16 @@ EXPORT_SYMBOL_GPL(mlx5_db_alloc);
 
 void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db)
 {
+	u32 db_per_page = PAGE_SIZE / cache_line_size();
 	mutex_lock(&dev->priv.pgdir_mutex);
 
 	__set_bit(db->index, db->u.pgdir->bitmap);
 
-	if (bitmap_full(db->u.pgdir->bitmap, MLX5_DB_PER_PAGE)) {
+	if (bitmap_full(db->u.pgdir->bitmap, db_per_page)) {
 		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
 				  db->u.pgdir->db_page, db->u.pgdir->db_dma);
 		list_del(&db->u.pgdir->list);
+		kfree(db->u.pgdir->bitmap);
 		kfree(db->u.pgdir);
 	}
 

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

@@ -85,6 +85,9 @@
 #define MLX5_MPWRQ_SMALL_PACKET_THRESHOLD	(128)
 
 #define MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ                 (64 * 1024)
+#define MLX5E_DEFAULT_LRO_TIMEOUT                       32
+#define MLX5E_LRO_TIMEOUT_ARR_SIZE                      4
+
 #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC      0x10
 #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE 0x3
 #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS      0x20
@@ -221,6 +224,7 @@ struct mlx5e_params {
 	struct ieee_ets ets;
 #endif
 	bool rx_am_enabled;
+	u32 lro_timeout;
 };
 
 struct mlx5e_tstamp {
@@ -888,5 +892,6 @@ int mlx5e_attach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev);
 void mlx5e_detach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev);
 struct rtnl_link_stats64 *
 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats);
+u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout);
 
 #endif /* __MLX5_EN_H__ */

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

@@ -1971,9 +1971,7 @@ static void mlx5e_build_tir_ctx_lro(void *tirc, struct mlx5e_priv *priv)
 	MLX5_SET(tirc, tirc, lro_max_ip_payload_size,
 		 (priv->params.lro_wqe_sz -
 		  ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
-	MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
-		 MLX5_CAP_ETH(priv->mdev,
-			      lro_timer_supported_periods[2]));
+	MLX5_SET(tirc, tirc, lro_timeout_period_usecs, priv->params.lro_timeout);
 }
 
 void mlx5e_build_tir_ctx_hash(void *tirc, struct mlx5e_priv *priv)
@@ -3401,6 +3399,18 @@ static void mlx5e_query_min_inline(struct mlx5_core_dev *mdev,
 	}
 }
 
+u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
+{
+	int i;
+
+	/* The supported periods are organized in ascending order */
+	for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
+		if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
+			break;
+
+	return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
+}
+
 static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev,
 					struct net_device *netdev,
 					const struct mlx5e_profile *profile,
@@ -3419,6 +3429,9 @@ static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev,
 	priv->profile                      = profile;
 	priv->ppriv                        = ppriv;
 
+	priv->params.lro_timeout =
+		mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
+
 	priv->params.log_sq_size = MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
 
 	/* set CQE compression */
@@ -4035,7 +4048,6 @@ void mlx5e_destroy_netdev(struct mlx5_core_dev *mdev, struct mlx5e_priv *priv)
 	const struct mlx5e_profile *profile = priv->profile;
 	struct net_device *netdev = priv->netdev;
 
-	unregister_netdev(netdev);
 	destroy_workqueue(priv->wq);
 	if (profile->cleanup)
 		profile->cleanup(priv);
@@ -4052,6 +4064,7 @@ static void mlx5e_remove(struct mlx5_core_dev *mdev, void *vpriv)
 	for (vport = 1; vport < total_vfs; vport++)
 		mlx5_eswitch_unregister_vport_rep(esw, vport);
 
+	unregister_netdev(priv->netdev);
 	mlx5e_detach(mdev, vpriv);
 	mlx5e_destroy_netdev(mdev, priv);
 }

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

@@ -457,6 +457,7 @@ void mlx5e_vport_rep_unload(struct mlx5_eswitch *esw,
 	struct mlx5e_priv *priv = rep->priv_data;
 	struct net_device *netdev = priv->netdev;
 
+	unregister_netdev(netdev);
 	mlx5e_detach_netdev(esw->dev, netdev);
 	mlx5e_destroy_netdev(esw->dev, priv);
 }

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

@@ -931,8 +931,8 @@ static void esw_vport_change_handler(struct work_struct *work)
 	mutex_unlock(&esw->state_lock);
 }
 
-static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw,
-					struct mlx5_vport *vport)
+static int esw_vport_enable_egress_acl(struct mlx5_eswitch *esw,
+				       struct mlx5_vport *vport)
 {
 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 	struct mlx5_flow_group *vlan_grp = NULL;
@@ -949,9 +949,11 @@ static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw,
 	int table_size = 2;
 	int err = 0;
 
-	if (!MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support) ||
-	    !IS_ERR_OR_NULL(vport->egress.acl))
-		return;
+	if (!MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
+		return -EOPNOTSUPP;
+
+	if (!IS_ERR_OR_NULL(vport->egress.acl))
+		return 0;
 
 	esw_debug(dev, "Create vport[%d] egress ACL log_max_size(%d)\n",
 		  vport->vport, MLX5_CAP_ESW_EGRESS_ACL(dev, log_max_ft_size));
@@ -959,12 +961,12 @@ static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw,
 	root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ESW_EGRESS);
 	if (!root_ns) {
 		esw_warn(dev, "Failed to get E-Switch egress flow namespace\n");
-		return;
+		return -EIO;
 	}
 
 	flow_group_in = mlx5_vzalloc(inlen);
 	if (!flow_group_in)
-		return;
+		return -ENOMEM;
 
 	acl = mlx5_create_vport_flow_table(root_ns, 0, table_size, 0, vport->vport);
 	if (IS_ERR(acl)) {
@@ -1009,6 +1011,7 @@ out:
 		mlx5_destroy_flow_group(vlan_grp);
 	if (err && !IS_ERR_OR_NULL(acl))
 		mlx5_destroy_flow_table(acl);
+	return err;
 }
 
 static void esw_vport_cleanup_egress_rules(struct mlx5_eswitch *esw,
@@ -1041,8 +1044,8 @@ static void esw_vport_disable_egress_acl(struct mlx5_eswitch *esw,
 	vport->egress.acl = NULL;
 }
 
-static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw,
-					 struct mlx5_vport *vport)
+static int esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw,
+					struct mlx5_vport *vport)
 {
 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 	struct mlx5_core_dev *dev = esw->dev;
@@ -1063,9 +1066,11 @@ static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw,
 	int table_size = 4;
 	int err = 0;
 
-	if (!MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support) ||
-	    !IS_ERR_OR_NULL(vport->ingress.acl))
-		return;
+	if (!MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support))
+		return -EOPNOTSUPP;
+
+	if (!IS_ERR_OR_NULL(vport->ingress.acl))
+		return 0;
 
 	esw_debug(dev, "Create vport[%d] ingress ACL log_max_size(%d)\n",
 		  vport->vport, MLX5_CAP_ESW_INGRESS_ACL(dev, log_max_ft_size));
@@ -1073,12 +1078,12 @@ static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw,
 	root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ESW_INGRESS);
 	if (!root_ns) {
 		esw_warn(dev, "Failed to get E-Switch ingress flow namespace\n");
-		return;
+		return -EIO;
 	}
 
 	flow_group_in = mlx5_vzalloc(inlen);
 	if (!flow_group_in)
-		return;
+		return -ENOMEM;
 
 	acl = mlx5_create_vport_flow_table(root_ns, 0, table_size, 0, vport->vport);
 	if (IS_ERR(acl)) {
@@ -1167,6 +1172,7 @@ out:
 	}
 
 	kvfree(flow_group_in);
+	return err;
 }
 
 static void esw_vport_cleanup_ingress_rules(struct mlx5_eswitch *esw,
@@ -1225,7 +1231,13 @@ static int esw_vport_ingress_config(struct mlx5_eswitch *esw,
 		return 0;
 	}
 
-	esw_vport_enable_ingress_acl(esw, vport);
+	err = esw_vport_enable_ingress_acl(esw, vport);
+	if (err) {
+		mlx5_core_warn(esw->dev,
+			       "failed to enable ingress acl (%d) on vport[%d]\n",
+			       err, vport->vport);
+		return err;
+	}
 
 	esw_debug(esw->dev,
 		  "vport[%d] configure ingress rules, vlan(%d) qos(%d)\n",
@@ -1299,7 +1311,13 @@ static int esw_vport_egress_config(struct mlx5_eswitch *esw,
 		return 0;
 	}
 
-	esw_vport_enable_egress_acl(esw, vport);
+	err = esw_vport_enable_egress_acl(esw, vport);
+	if (err) {
+		mlx5_core_warn(esw->dev,
+			       "failed to enable egress acl (%d) on vport[%d]\n",
+			       err, vport->vport);
+		return err;
+	}
 
 	esw_debug(esw->dev,
 		  "vport[%d] configure egress rules, vlan(%d) qos(%d)\n",

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

@@ -436,6 +436,9 @@ static void del_flow_group(struct fs_node *node)
 	fs_get_obj(ft, fg->node.parent);
 	dev = get_dev(&ft->node);
 
+	if (ft->autogroup.active)
+		ft->autogroup.num_groups--;
+
 	if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 		mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 			       fg->id, ft->id);
@@ -879,7 +882,7 @@ static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *
 	tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 	tree_add_node(&fg->node, &ft->node);
 	/* Add node to group list */
-	list_add(&fg->node.list, ft->node.children.prev);
+	list_add(&fg->node.list, prev_fg);
 
 	return fg;
 }
@@ -893,7 +896,7 @@ struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 		return ERR_PTR(-EPERM);
 
 	lock_ref_node(&ft->node);
-	fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
+	fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
 	unlock_ref_node(&ft->node);
 
 	return fg;
@@ -1012,7 +1015,7 @@ static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 						u32 *match_criteria)
 {
 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
-	struct list_head *prev = &ft->node.children;
+	struct list_head *prev = ft->node.children.prev;
 	unsigned int candidate_index = 0;
 	struct mlx5_flow_group *fg;
 	void *match_criteria_addr;

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

@@ -218,6 +218,7 @@ struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)
 		goto err_out;
 
 	if (aging) {
+		counter->cache.lastuse = jiffies;
 		counter->aging = true;
 
 		spin_lock(&fc_stats->addlist_lock);

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

@@ -61,10 +61,15 @@ enum {
 enum {
 	MLX5_NIC_IFC_FULL		= 0,
 	MLX5_NIC_IFC_DISABLED		= 1,
-	MLX5_NIC_IFC_NO_DRAM_NIC	= 2
+	MLX5_NIC_IFC_NO_DRAM_NIC	= 2,
+	MLX5_NIC_IFC_INVALID		= 3
 };
 
-static u8 get_nic_interface(struct mlx5_core_dev *dev)
+enum {
+	MLX5_DROP_NEW_HEALTH_WORK,
+};
+
+static u8 get_nic_state(struct mlx5_core_dev *dev)
 {
 	return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 3;
 }
@@ -97,7 +102,7 @@ static int in_fatal(struct mlx5_core_dev *dev)
 	struct mlx5_core_health *health = &dev->priv.health;
 	struct health_buffer __iomem *h = health->health;
 
-	if (get_nic_interface(dev) == MLX5_NIC_IFC_DISABLED)
+	if (get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
 		return 1;
 
 	if (ioread32be(&h->fw_ver) == 0xffffffff)
@@ -127,7 +132,7 @@ unlock:
 
 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
 {
-	u8 nic_interface = get_nic_interface(dev);
+	u8 nic_interface = get_nic_state(dev);
 
 	switch (nic_interface) {
 	case MLX5_NIC_IFC_FULL:
@@ -149,8 +154,34 @@ static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
 	mlx5_disable_device(dev);
 }
 
+static void health_recover(struct work_struct *work)
+{
+	struct mlx5_core_health *health;
+	struct delayed_work *dwork;
+	struct mlx5_core_dev *dev;
+	struct mlx5_priv *priv;
+	u8 nic_state;
+
+	dwork = container_of(work, struct delayed_work, work);
+	health = container_of(dwork, struct mlx5_core_health, recover_work);
+	priv = container_of(health, struct mlx5_priv, health);
+	dev = container_of(priv, struct mlx5_core_dev, priv);
+
+	nic_state = get_nic_state(dev);
+	if (nic_state == MLX5_NIC_IFC_INVALID) {
+		dev_err(&dev->pdev->dev, "health recovery flow aborted since the nic state is invalid\n");
+		return;
+	}
+
+	dev_err(&dev->pdev->dev, "starting health recovery flow\n");
+	mlx5_recover_device(dev);
+}
+
+/* How much time to wait until health resetting the driver (in msecs) */
+#define MLX5_RECOVERY_DELAY_MSECS 60000
 static void health_care(struct work_struct *work)
 {
+	unsigned long recover_delay = msecs_to_jiffies(MLX5_RECOVERY_DELAY_MSECS);
 	struct mlx5_core_health *health;
 	struct mlx5_core_dev *dev;
 	struct mlx5_priv *priv;
@@ -160,6 +191,14 @@ static void health_care(struct work_struct *work)
 	dev = container_of(priv, struct mlx5_core_dev, priv);
 	mlx5_core_warn(dev, "handling bad device here\n");
 	mlx5_handle_bad_state(dev);
+
+	spin_lock(&health->wq_lock);
+	if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
+		schedule_delayed_work(&health->recover_work, recover_delay);
+	else
+		dev_err(&dev->pdev->dev,
+			"new health works are not permitted at this stage\n");
+	spin_unlock(&health->wq_lock);
 }
 
 static const char *hsynd_str(u8 synd)
@@ -272,7 +311,13 @@ static void poll_health(unsigned long data)
 	if (in_fatal(dev) && !health->sick) {
 		health->sick = true;
 		print_health_info(dev);
-		schedule_work(&health->work);
+		spin_lock(&health->wq_lock);
+		if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
+			queue_work(health->wq, &health->work);
+		else
+			dev_err(&dev->pdev->dev,
+				"new health works are not permitted at this stage\n");
+		spin_unlock(&health->wq_lock);
 	}
 }
 
@@ -281,6 +326,8 @@ void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 	struct mlx5_core_health *health = &dev->priv.health;
 
 	init_timer(&health->timer);
+	health->sick = 0;
+	clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
 	health->health = &dev->iseg->health;
 	health->health_counter = &dev->iseg->health_counter;
 
@@ -297,11 +344,22 @@ void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
 	del_timer_sync(&health->timer);
 }
 
+void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
+{
+	struct mlx5_core_health *health = &dev->priv.health;
+
+	spin_lock(&health->wq_lock);
+	set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
+	spin_unlock(&health->wq_lock);
+	cancel_delayed_work_sync(&health->recover_work);
+	cancel_work_sync(&health->work);
+}
+
 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
 {
 	struct mlx5_core_health *health = &dev->priv.health;
 
-	flush_work(&health->work);
+	destroy_workqueue(health->wq);
 }
 
 int mlx5_health_init(struct mlx5_core_dev *dev)
@@ -316,9 +374,13 @@ int mlx5_health_init(struct mlx5_core_dev *dev)
 
 	strcpy(name, "mlx5_health");
 	strcat(name, dev_name(&dev->pdev->dev));
+	health->wq = create_singlethread_workqueue(name);
 	kfree(name);
-
+	if (!health->wq)
+		return -ENOMEM;
+	spin_lock_init(&health->wq_lock);
 	INIT_WORK(&health->work, health_care);
+	INIT_DELAYED_WORK(&health->recover_work, health_recover);
 
 	return 0;
 }

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

@@ -844,12 +844,6 @@ static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
 	struct pci_dev *pdev = dev->pdev;
 	int err;
 
-	err = mlx5_query_hca_caps(dev);
-	if (err) {
-		dev_err(&pdev->dev, "query hca failed\n");
-		goto out;
-	}
-
 	err = mlx5_query_board_id(dev);
 	if (err) {
 		dev_err(&pdev->dev, "query board id failed\n");
@@ -1023,6 +1017,12 @@ static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
 
 	mlx5_start_health_poll(dev);
 
+	err = mlx5_query_hca_caps(dev);
+	if (err) {
+		dev_err(&pdev->dev, "query hca failed\n");
+		goto err_stop_poll;
+	}
+
 	if (boot && mlx5_init_once(dev, priv)) {
 		dev_err(&pdev->dev, "sw objs init failed\n");
 		goto err_stop_poll;
@@ -1313,10 +1313,16 @@ static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
 	struct mlx5_priv *priv = &dev->priv;
 
 	dev_info(&pdev->dev, "%s was called\n", __func__);
+
 	mlx5_enter_error_state(dev);
 	mlx5_unload_one(dev, priv, false);
-	pci_save_state(pdev);
-	mlx5_pci_disable_device(dev);
+	/* In case of kernel call save the pci state and drain health wq */
+	if (state) {
+		pci_save_state(pdev);
+		mlx5_drain_health_wq(dev);
+		mlx5_pci_disable_device(dev);
+	}
+
 	return state == pci_channel_io_perm_failure ?
 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
 }
@@ -1373,11 +1379,6 @@ static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
 	return PCI_ERS_RESULT_RECOVERED;
 }
 
-void mlx5_disable_device(struct mlx5_core_dev *dev)
-{
-	mlx5_pci_err_detected(dev->pdev, 0);
-}
-
 static void mlx5_pci_resume(struct pci_dev *pdev)
 {
 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
@@ -1427,6 +1428,18 @@ static const struct pci_device_id mlx5_core_pci_table[] = {
 
 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
 
+void mlx5_disable_device(struct mlx5_core_dev *dev)
+{
+	mlx5_pci_err_detected(dev->pdev, 0);
+}
+
+void mlx5_recover_device(struct mlx5_core_dev *dev)
+{
+	mlx5_pci_disable_device(dev);
+	if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
+		mlx5_pci_resume(dev->pdev);
+}
+
 static struct pci_driver mlx5_core_driver = {
 	.name           = DRIVER_NAME,
 	.id_table       = mlx5_core_pci_table,

drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h

@@ -83,6 +83,7 @@ void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
 		     unsigned long param);
 void mlx5_enter_error_state(struct mlx5_core_dev *dev);
 void mlx5_disable_device(struct mlx5_core_dev *dev);
+void mlx5_recover_device(struct mlx5_core_dev *dev);
 int mlx5_sriov_init(struct mlx5_core_dev *dev);
 void mlx5_sriov_cleanup(struct mlx5_core_dev *dev);
 int mlx5_sriov_attach(struct mlx5_core_dev *dev);

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

@@ -209,6 +209,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr)
 static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
 {
 	struct page *page;
+	u64 zero_addr = 1;
 	u64 addr;
 	int err;
 	int nid = dev_to_node(&dev->pdev->dev);
@@ -218,26 +219,35 @@ static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
 		mlx5_core_warn(dev, "failed to allocate page\n");
 		return -ENOMEM;
 	}
+map:
 	addr = dma_map_page(&dev->pdev->dev, page, 0,
 			    PAGE_SIZE, DMA_BIDIRECTIONAL);
 	if (dma_mapping_error(&dev->pdev->dev, addr)) {
 		mlx5_core_warn(dev, "failed dma mapping page\n");
 		err = -ENOMEM;
-		goto out_alloc;
+		goto err_mapping;
 	}
+
+	/* Firmware doesn't support page with physical address 0 */
+	if (addr == 0) {
+		zero_addr = addr;
+		goto map;
+	}
+
 	err = insert_page(dev, addr, page, func_id);
 	if (err) {
 		mlx5_core_err(dev, "failed to track allocated page\n");
-		goto out_mapping;
+		dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE,
+			       DMA_BIDIRECTIONAL);
 	}
 
-	return 0;
-
-out_mapping:
-	dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
+err_mapping:
+	if (err)
+		__free_page(page);
 
-out_alloc:
-	__free_page(page);
+	if (zero_addr == 0)
+		dma_unmap_page(&dev->pdev->dev, zero_addr, PAGE_SIZE,
+			       DMA_BIDIRECTIONAL);
 
 	return err;
 }

include/linux/mlx5/driver.h

@@ -418,8 +418,12 @@ struct mlx5_core_health {
 	u32				prev;
 	int				miss_counter;
 	bool				sick;
+	/* wq spinlock to synchronize draining */
+	spinlock_t			wq_lock;
 	struct workqueue_struct	       *wq;
+	unsigned long			flags;
 	struct work_struct		work;
+	struct delayed_work		recover_work;
 };
 
 struct mlx5_cq_table {
@@ -625,10 +629,6 @@ struct mlx5_db {
 	int			index;
 };
 
-enum {
-	MLX5_DB_PER_PAGE = PAGE_SIZE / L1_CACHE_BYTES,
-};
-
 enum {
 	MLX5_COMP_EQ_SIZE = 1024,
 };
@@ -638,13 +638,6 @@ enum {
 	MLX5_PTYS_EN = 1 << 2,
 };
 
-struct mlx5_db_pgdir {
-	struct list_head	list;
-	DECLARE_BITMAP(bitmap, MLX5_DB_PER_PAGE);
-	__be32		       *db_page;
-	dma_addr_t		db_dma;
-};
-
 typedef void (*mlx5_cmd_cbk_t)(int status, void *context);
 
 struct mlx5_cmd_work_ent {
@@ -789,6 +782,7 @@ void mlx5_health_cleanup(struct mlx5_core_dev *dev);
 int mlx5_health_init(struct mlx5_core_dev *dev);
 void mlx5_start_health_poll(struct mlx5_core_dev *dev);
 void mlx5_stop_health_poll(struct mlx5_core_dev *dev);
+void mlx5_drain_health_wq(struct mlx5_core_dev *dev);
 int mlx5_buf_alloc_node(struct mlx5_core_dev *dev, int size,
 			struct mlx5_buf *buf, int node);
 int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, struct mlx5_buf *buf);