Merge tag 'mlx5-fixes-2020-09-30' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

From: Saeed Mahameed <saeedm@nvidia.com>

====================
This series introduces some fixes to mlx5 driver.

v1->v2:
 - Patch #1 Don't return while mutex is held. (Dave)

v2->v3:
 - Drop patch #1, will consider a better approach (Jakub)
 - use cpu_relax() instead of cond_resched() (Jakub)
 - while(i--) to reveres a loop (Jakub)
 - Drop old mellanox email sign-off and change the committer email
   (Jakub)

Please pull and let me know if there is any problem.

For -stable v4.15
 ('net/mlx5e: Fix VLAN cleanup flow')
 ('net/mlx5e: Fix VLAN create flow')

For -stable v4.16
 ('net/mlx5: Fix request_irqs error flow')

For -stable v5.4
 ('net/mlx5e: Add resiliency in Striding RQ mode for packets larger than MTU')
 ('net/mlx5: Avoid possible free of command entry while timeout comp handler')

For -stable v5.7
 ('net/mlx5e: Fix return status when setting unsupported FEC mode')

For -stable v5.8
 ('net/mlx5e: Fix race condition on nhe->n pointer in neigh update')
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

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

@@ -69,12 +69,10 @@ enum {
 	MLX5_CMD_DELIVERY_STAT_CMD_DESCR_ERR		= 0x10,
 };
 
-static struct mlx5_cmd_work_ent *alloc_cmd(struct mlx5_cmd *cmd,
-					   struct mlx5_cmd_msg *in,
-					   struct mlx5_cmd_msg *out,
-					   void *uout, int uout_size,
-					   mlx5_cmd_cbk_t cbk,
-					   void *context, int page_queue)
+static struct mlx5_cmd_work_ent *
+cmd_alloc_ent(struct mlx5_cmd *cmd, struct mlx5_cmd_msg *in,
+	      struct mlx5_cmd_msg *out, void *uout, int uout_size,
+	      mlx5_cmd_cbk_t cbk, void *context, int page_queue)
 {
 	gfp_t alloc_flags = cbk ? GFP_ATOMIC : GFP_KERNEL;
 	struct mlx5_cmd_work_ent *ent;
@@ -83,6 +81,7 @@ static struct mlx5_cmd_work_ent *alloc_cmd(struct mlx5_cmd *cmd,
 	if (!ent)
 		return ERR_PTR(-ENOMEM);
 
+	ent->idx	= -EINVAL;
 	ent->in		= in;
 	ent->out	= out;
 	ent->uout	= uout;
@@ -91,10 +90,16 @@ static struct mlx5_cmd_work_ent *alloc_cmd(struct mlx5_cmd *cmd,
 	ent->context	= context;
 	ent->cmd	= cmd;
 	ent->page_queue = page_queue;
+	refcount_set(&ent->refcnt, 1);
 
 	return ent;
 }
 
+static void cmd_free_ent(struct mlx5_cmd_work_ent *ent)
+{
+	kfree(ent);
+}
+
 static u8 alloc_token(struct mlx5_cmd *cmd)
 {
 	u8 token;
@@ -109,7 +114,7 @@ static u8 alloc_token(struct mlx5_cmd *cmd)
 	return token;
 }
 
-static int alloc_ent(struct mlx5_cmd *cmd)
+static int cmd_alloc_index(struct mlx5_cmd *cmd)
 {
 	unsigned long flags;
 	int ret;
@@ -123,7 +128,7 @@ static int alloc_ent(struct mlx5_cmd *cmd)
 	return ret < cmd->max_reg_cmds ? ret : -ENOMEM;
 }
 
-static void free_ent(struct mlx5_cmd *cmd, int idx)
+static void cmd_free_index(struct mlx5_cmd *cmd, int idx)
 {
 	unsigned long flags;
 
@@ -132,6 +137,22 @@ static void free_ent(struct mlx5_cmd *cmd, int idx)
 	spin_unlock_irqrestore(&cmd->alloc_lock, flags);
 }
 
+static void cmd_ent_get(struct mlx5_cmd_work_ent *ent)
+{
+	refcount_inc(&ent->refcnt);
+}
+
+static void cmd_ent_put(struct mlx5_cmd_work_ent *ent)
+{
+	if (!refcount_dec_and_test(&ent->refcnt))
+		return;
+
+	if (ent->idx >= 0)
+		cmd_free_index(ent->cmd, ent->idx);
+
+	cmd_free_ent(ent);
+}
+
 static struct mlx5_cmd_layout *get_inst(struct mlx5_cmd *cmd, int idx)
 {
 	return cmd->cmd_buf + (idx << cmd->log_stride);
@@ -219,11 +240,6 @@ static void poll_timeout(struct mlx5_cmd_work_ent *ent)
 	ent->ret = -ETIMEDOUT;
 }
 
-static void free_cmd(struct mlx5_cmd_work_ent *ent)
-{
-	kfree(ent);
-}
-
 static int verify_signature(struct mlx5_cmd_work_ent *ent)
 {
 	struct mlx5_cmd_mailbox *next = ent->out->next;
@@ -837,11 +853,22 @@ static void cb_timeout_handler(struct work_struct *work)
 	struct mlx5_core_dev *dev = container_of(ent->cmd, struct mlx5_core_dev,
 						 cmd);
 
+	mlx5_cmd_eq_recover(dev);
+
+	/* Maybe got handled by eq recover ? */
+	if (!test_bit(MLX5_CMD_ENT_STATE_PENDING_COMP, &ent->state)) {
+		mlx5_core_warn(dev, "cmd[%d]: %s(0x%x) Async, recovered after timeout\n", ent->idx,
+			       mlx5_command_str(msg_to_opcode(ent->in)), msg_to_opcode(ent->in));
+		goto out; /* phew, already handled */
+	}
+
 	ent->ret = -ETIMEDOUT;
-	mlx5_core_warn(dev, "%s(0x%x) timeout. Will cause a leak of a command resource\n",
-		       mlx5_command_str(msg_to_opcode(ent->in)),
-		       msg_to_opcode(ent->in));
+	mlx5_core_warn(dev, "cmd[%d]: %s(0x%x) Async, timeout. Will cause a leak of a command resource\n",
+		       ent->idx, mlx5_command_str(msg_to_opcode(ent->in)), msg_to_opcode(ent->in));
 	mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
+
+out:
+	cmd_ent_put(ent); /* for the cmd_ent_get() took on schedule delayed work */
 }
 
 static void free_msg(struct mlx5_core_dev *dev, struct mlx5_cmd_msg *msg);
@@ -856,6 +883,32 @@ static bool opcode_allowed(struct mlx5_cmd *cmd, u16 opcode)
 	return cmd->allowed_opcode == opcode;
 }
 
+static int cmd_alloc_index_retry(struct mlx5_cmd *cmd)
+{
+	unsigned long alloc_end = jiffies + msecs_to_jiffies(1000);
+	int idx;
+
+retry:
+	idx = cmd_alloc_index(cmd);
+	if (idx < 0 && time_before(jiffies, alloc_end)) {
+		/* Index allocation can fail on heavy load of commands. This is a temporary
+		 * situation as the current command already holds the semaphore, meaning that
+		 * another command completion is being handled and it is expected to release
+		 * the entry index soon.
+		 */
+		cpu_relax();
+		goto retry;
+	}
+	return idx;
+}
+
+bool mlx5_cmd_is_down(struct mlx5_core_dev *dev)
+{
+	return pci_channel_offline(dev->pdev) ||
+	       dev->cmd.state != MLX5_CMDIF_STATE_UP ||
+	       dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR;
+}
+
 static void cmd_work_handler(struct work_struct *work)
 {
 	struct mlx5_cmd_work_ent *ent = container_of(work, struct mlx5_cmd_work_ent, work);
@@ -873,14 +926,14 @@ static void cmd_work_handler(struct work_struct *work)
 	sem = ent->page_queue ? &cmd->pages_sem : &cmd->sem;
 	down(sem);
 	if (!ent->page_queue) {
-		alloc_ret = alloc_ent(cmd);
+		alloc_ret = cmd_alloc_index_retry(cmd);
 		if (alloc_ret < 0) {
 			mlx5_core_err_rl(dev, "failed to allocate command entry\n");
 			if (ent->callback) {
 				ent->callback(-EAGAIN, ent->context);
 				mlx5_free_cmd_msg(dev, ent->out);
 				free_msg(dev, ent->in);
-				free_cmd(ent);
+				cmd_ent_put(ent);
 			} else {
 				ent->ret = -EAGAIN;
 				complete(&ent->done);
@@ -916,15 +969,12 @@ static void cmd_work_handler(struct work_struct *work)
 	ent->ts1 = ktime_get_ns();
 	cmd_mode = cmd->mode;
 
-	if (ent->callback)
-		schedule_delayed_work(&ent->cb_timeout_work, cb_timeout);
+	if (ent->callback && schedule_delayed_work(&ent->cb_timeout_work, cb_timeout))
+		cmd_ent_get(ent);
 	set_bit(MLX5_CMD_ENT_STATE_PENDING_COMP, &ent->state);
 
 	/* Skip sending command to fw if internal error */
-	if (pci_channel_offline(dev->pdev) ||
-	    dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR ||
-	    cmd->state != MLX5_CMDIF_STATE_UP ||
-	    !opcode_allowed(&dev->cmd, ent->op)) {
+	if (mlx5_cmd_is_down(dev) || !opcode_allowed(&dev->cmd, ent->op)) {
 		u8 status = 0;
 		u32 drv_synd;
 
@@ -933,13 +983,10 @@ static void cmd_work_handler(struct work_struct *work)
 		MLX5_SET(mbox_out, ent->out, syndrome, drv_synd);
 
 		mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
-		/* no doorbell, no need to keep the entry */
-		free_ent(cmd, ent->idx);
-		if (ent->callback)
-			free_cmd(ent);
 		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();
@@ -983,6 +1030,35 @@ static const char *deliv_status_to_str(u8 status)
 	}
 }
 
+enum {
+	MLX5_CMD_TIMEOUT_RECOVER_MSEC   = 5 * 1000,
+};
+
+static void wait_func_handle_exec_timeout(struct mlx5_core_dev *dev,
+					  struct mlx5_cmd_work_ent *ent)
+{
+	unsigned long timeout = msecs_to_jiffies(MLX5_CMD_TIMEOUT_RECOVER_MSEC);
+
+	mlx5_cmd_eq_recover(dev);
+
+	/* Re-wait on the ent->done after executing the recovery flow. If the
+	 * recovery flow (or any other recovery flow running simultaneously)
+	 * has recovered an EQE, it should cause the entry to be completed by
+	 * the command interface.
+	 */
+	if (wait_for_completion_timeout(&ent->done, timeout)) {
+		mlx5_core_warn(dev, "cmd[%d]: %s(0x%x) recovered after timeout\n", ent->idx,
+			       mlx5_command_str(msg_to_opcode(ent->in)), msg_to_opcode(ent->in));
+		return;
+	}
+
+	mlx5_core_warn(dev, "cmd[%d]: %s(0x%x) No done completion\n", ent->idx,
+		       mlx5_command_str(msg_to_opcode(ent->in)), msg_to_opcode(ent->in));
+
+	ent->ret = -ETIMEDOUT;
+	mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
+}
+
 static int wait_func(struct mlx5_core_dev *dev, struct mlx5_cmd_work_ent *ent)
 {
 	unsigned long timeout = msecs_to_jiffies(MLX5_CMD_TIMEOUT_MSEC);
@@ -994,12 +1070,10 @@ static int wait_func(struct mlx5_core_dev *dev, struct mlx5_cmd_work_ent *ent)
 		ent->ret = -ECANCELED;
 		goto out_err;
 	}
-	if (cmd->mode == CMD_MODE_POLLING || ent->polling) {
+	if (cmd->mode == CMD_MODE_POLLING || ent->polling)
 		wait_for_completion(&ent->done);
-	} else if (!wait_for_completion_timeout(&ent->done, timeout)) {
-		ent->ret = -ETIMEDOUT;
-		mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
-	}
+	else if (!wait_for_completion_timeout(&ent->done, timeout))
+		wait_func_handle_exec_timeout(dev, ent);
 
 out_err:
 	err = ent->ret;
@@ -1039,11 +1113,16 @@ static int mlx5_cmd_invoke(struct mlx5_core_dev *dev, struct mlx5_cmd_msg *in,
 	if (callback && page_queue)
 		return -EINVAL;
 
-	ent = alloc_cmd(cmd, in, out, uout, uout_size, callback, context,
-			page_queue);
+	ent = cmd_alloc_ent(cmd, in, out, uout, uout_size,
+			    callback, context, page_queue);
 	if (IS_ERR(ent))
 		return PTR_ERR(ent);
 
+	/* put for this ent is when consumed, depending on the use case
+	 * 1) (!callback) blocking flow: by caller after wait_func completes
+	 * 2) (callback) flow: by mlx5_cmd_comp_handler() when ent is handled
+	 */
+
 	ent->token = token;
 	ent->polling = force_polling;
 
@@ -1062,12 +1141,10 @@ static int mlx5_cmd_invoke(struct mlx5_core_dev *dev, struct mlx5_cmd_msg *in,
 	}
 
 	if (callback)
-		goto out;
+		goto out; /* mlx5_cmd_comp_handler() will put(ent) */
 
 	err = wait_func(dev, ent);
-	if (err == -ETIMEDOUT)
-		goto out;
-	if (err == -ECANCELED)
+	if (err == -ETIMEDOUT || err == -ECANCELED)
 		goto out_free;
 
 	ds = ent->ts2 - ent->ts1;
@@ -1085,7 +1162,7 @@ static int mlx5_cmd_invoke(struct mlx5_core_dev *dev, struct mlx5_cmd_msg *in,
 	*status = ent->status;
 
 out_free:
-	free_cmd(ent);
+	cmd_ent_put(ent);
 out:
 	return err;
 }
@@ -1516,14 +1593,19 @@ static void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool force
 				if (!forced) {
 					mlx5_core_err(dev, "Command completion arrived after timeout (entry idx = %d).\n",
 						      ent->idx);
-					free_ent(cmd, ent->idx);
-					free_cmd(ent);
+					cmd_ent_put(ent);
 				}
 				continue;
 			}
 
-			if (ent->callback)
-				cancel_delayed_work(&ent->cb_timeout_work);
+			if (ent->callback && cancel_delayed_work(&ent->cb_timeout_work))
+				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)
+				cmd_ent_put(ent);
+
 			if (ent->page_queue)
 				sem = &cmd->pages_sem;
 			else
@@ -1545,10 +1627,6 @@ static void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool force
 					      ent->ret, deliv_status_to_str(ent->status), ent->status);
 			}
 
-			/* only real completion will free the entry slot */
-			if (!forced)
-				free_ent(cmd, ent->idx);
-
 			if (ent->callback) {
 				ds = ent->ts2 - ent->ts1;
 				if (ent->op < MLX5_CMD_OP_MAX) {
@@ -1576,10 +1654,13 @@ static void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool force
 				free_msg(dev, ent->in);
 
 				err = err ? err : ent->status;
-				if (!forced)
-					free_cmd(ent);
+				/* final consumer is done, release ent */
+				cmd_ent_put(ent);
 				callback(err, context);
 			} else {
+				/* release wait_func() so mlx5_cmd_invoke()
+				 * can make the final ent_put()
+				 */
 				complete(&ent->done);
 			}
 			up(sem);
@@ -1589,8 +1670,11 @@ static void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool force
 
 void mlx5_cmd_trigger_completions(struct mlx5_core_dev *dev)
 {
+	struct mlx5_cmd *cmd = &dev->cmd;
+	unsigned long bitmask;
 	unsigned long flags;
 	u64 vector;
+	int i;
 
 	/* wait for pending handlers to complete */
 	mlx5_eq_synchronize_cmd_irq(dev);
@@ -1599,11 +1683,20 @@ void mlx5_cmd_trigger_completions(struct mlx5_core_dev *dev)
 	if (!vector)
 		goto no_trig;
 
+	bitmask = vector;
+	/* we must increment the allocated entries refcount before triggering the completions
+	 * to guarantee pending commands will not get freed in the meanwhile.
+	 * For that reason, it also has to be done inside the alloc_lock.
+	 */
+	for_each_set_bit(i, &bitmask, (1 << cmd->log_sz))
+		cmd_ent_get(cmd->ent_arr[i]);
 	vector |= MLX5_TRIGGERED_CMD_COMP;
 	spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
 
 	mlx5_core_dbg(dev, "vector 0x%llx\n", vector);
 	mlx5_cmd_comp_handler(dev, vector, true);
+	for_each_set_bit(i, &bitmask, (1 << cmd->log_sz))
+		cmd_ent_put(cmd->ent_arr[i]);
 	return;
 
 no_trig:
@@ -1711,10 +1804,7 @@ static int cmd_exec(struct mlx5_core_dev *dev, void *in, int in_size, void *out,
 	u8 token;
 
 	opcode = MLX5_GET(mbox_in, in, opcode);
-	if (pci_channel_offline(dev->pdev) ||
-	    dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR ||
-	    dev->cmd.state != MLX5_CMDIF_STATE_UP ||
-	    !opcode_allowed(&dev->cmd, opcode)) {
+	if (mlx5_cmd_is_down(dev) || !opcode_allowed(&dev->cmd, opcode)) {
 		err = mlx5_internal_err_ret_value(dev, opcode, &drv_synd, &status);
 		MLX5_SET(mbox_out, out, status, status);
 		MLX5_SET(mbox_out, out, syndrome, drv_synd);

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

@@ -91,7 +91,12 @@ struct page_pool;
 #define MLX5_MPWRQ_PAGES_PER_WQE		BIT(MLX5_MPWRQ_WQE_PAGE_ORDER)
 
 #define MLX5_MTT_OCTW(npages) (ALIGN(npages, 8) / 2)
-#define MLX5E_REQUIRED_WQE_MTTS		(ALIGN(MLX5_MPWRQ_PAGES_PER_WQE, 8))
+/* Add another page to MLX5E_REQUIRED_WQE_MTTS as a buffer between
+ * WQEs, This page will absorb write overflow by the hardware, when
+ * receiving packets larger than MTU. These oversize packets are
+ * dropped by the driver at a later stage.
+ */
+#define MLX5E_REQUIRED_WQE_MTTS		(ALIGN(MLX5_MPWRQ_PAGES_PER_WQE + 1, 8))
 #define MLX5E_LOG_ALIGNED_MPWQE_PPW	(ilog2(MLX5E_REQUIRED_WQE_MTTS))
 #define MLX5E_REQUIRED_MTTS(wqes)	(wqes * MLX5E_REQUIRED_WQE_MTTS)
 #define MLX5E_MAX_RQ_NUM_MTTS	\
@@ -617,6 +622,7 @@ struct mlx5e_rq {
 	u32                    rqn;
 	struct mlx5_core_dev  *mdev;
 	struct mlx5_core_mkey  umr_mkey;
+	struct mlx5e_dma_info  wqe_overflow;
 
 	/* XDP read-mostly */
 	struct xdp_rxq_info    xdp_rxq;

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

@@ -569,6 +569,9 @@ int mlx5e_set_fec_mode(struct mlx5_core_dev *dev, u16 fec_policy)
 	if (fec_policy >= (1 << MLX5E_FEC_LLRS_272_257_1) && !fec_50g_per_lane)
 		return -EOPNOTSUPP;
 
+	if (fec_policy && !mlx5e_fec_in_caps(dev, fec_policy))
+		return -EOPNOTSUPP;
+
 	MLX5_SET(pplm_reg, in, local_port, 1);
 	err = mlx5_core_access_reg(dev, in, sz, out, sz, MLX5_REG_PPLM, 0, 0);
 	if (err)

drivers/net/ethernet/mellanox/mlx5/core/en/rep/neigh.c

@@ -110,11 +110,25 @@ static void mlx5e_rep_neigh_stats_work(struct work_struct *work)
 	rtnl_unlock();
 }
 
+struct neigh_update_work {
+	struct work_struct work;
+	struct neighbour *n;
+	struct mlx5e_neigh_hash_entry *nhe;
+};
+
+static void mlx5e_release_neigh_update_work(struct neigh_update_work *update_work)
+{
+	neigh_release(update_work->n);
+	mlx5e_rep_neigh_entry_release(update_work->nhe);
+	kfree(update_work);
+}
+
 static void mlx5e_rep_neigh_update(struct work_struct *work)
 {
-	struct mlx5e_neigh_hash_entry *nhe =
-		container_of(work, struct mlx5e_neigh_hash_entry, neigh_update_work);
-	struct neighbour *n = nhe->n;
+	struct neigh_update_work *update_work = container_of(work, struct neigh_update_work,
+							     work);
+	struct mlx5e_neigh_hash_entry *nhe = update_work->nhe;
+	struct neighbour *n = update_work->n;
 	struct mlx5e_encap_entry *e;
 	unsigned char ha[ETH_ALEN];
 	struct mlx5e_priv *priv;
@@ -146,30 +160,42 @@ static void mlx5e_rep_neigh_update(struct work_struct *work)
 		mlx5e_rep_update_flows(priv, e, neigh_connected, ha);
 		mlx5e_encap_put(priv, e);
 	}
-	mlx5e_rep_neigh_entry_release(nhe);
 	rtnl_unlock();
-	neigh_release(n);
+	mlx5e_release_neigh_update_work(update_work);
 }
 
-static void mlx5e_rep_queue_neigh_update_work(struct mlx5e_priv *priv,
-					      struct mlx5e_neigh_hash_entry *nhe,
-					      struct neighbour *n)
+static struct neigh_update_work *mlx5e_alloc_neigh_update_work(struct mlx5e_priv *priv,
+							       struct neighbour *n)
 {
-	/* Take a reference to ensure the neighbour and mlx5 encap
-	 * entry won't be destructed until we drop the reference in
-	 * delayed work.
-	 */
-	neigh_hold(n);
+	struct neigh_update_work *update_work;
+	struct mlx5e_neigh_hash_entry *nhe;
+	struct mlx5e_neigh m_neigh = {};
 
-	/* This assignment is valid as long as the the neigh reference
-	 * is taken
-	 */
-	nhe->n = n;
+	update_work = kzalloc(sizeof(*update_work), GFP_ATOMIC);
+	if (WARN_ON(!update_work))
+		return NULL;
 
-	if (!queue_work(priv->wq, &nhe->neigh_update_work)) {
-		mlx5e_rep_neigh_entry_release(nhe);
-		neigh_release(n);
+	m_neigh.dev = n->dev;
+	m_neigh.family = n->ops->family;
+	memcpy(&m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
+
+	/* Obtain reference to nhe as last step in order not to release it in
+	 * atomic context.
+	 */
+	rcu_read_lock();
+	nhe = mlx5e_rep_neigh_entry_lookup(priv, &m_neigh);
+	rcu_read_unlock();
+	if (!nhe) {
+		kfree(update_work);
+		return NULL;
 	}
+
+	INIT_WORK(&update_work->work, mlx5e_rep_neigh_update);
+	neigh_hold(n);
+	update_work->n = n;
+	update_work->nhe = nhe;
+
+	return update_work;
 }
 
 static int mlx5e_rep_netevent_event(struct notifier_block *nb,
@@ -181,7 +207,7 @@ static int mlx5e_rep_netevent_event(struct notifier_block *nb,
 	struct net_device *netdev = rpriv->netdev;
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	struct mlx5e_neigh_hash_entry *nhe = NULL;
-	struct mlx5e_neigh m_neigh = {};
+	struct neigh_update_work *update_work;
 	struct neigh_parms *p;
 	struct neighbour *n;
 	bool found = false;
@@ -196,17 +222,11 @@ static int mlx5e_rep_netevent_event(struct notifier_block *nb,
 #endif
 			return NOTIFY_DONE;
 
-		m_neigh.dev = n->dev;
-		m_neigh.family = n->ops->family;
-		memcpy(&m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
-
-		rcu_read_lock();
-		nhe = mlx5e_rep_neigh_entry_lookup(priv, &m_neigh);
-		rcu_read_unlock();
-		if (!nhe)
+		update_work = mlx5e_alloc_neigh_update_work(priv, n);
+		if (!update_work)
 			return NOTIFY_DONE;
 
-		mlx5e_rep_queue_neigh_update_work(priv, nhe, n);
+		queue_work(priv->wq, &update_work->work);
 		break;
 
 	case NETEVENT_DELAY_PROBE_TIME_UPDATE:
@@ -352,7 +372,6 @@ int mlx5e_rep_neigh_entry_create(struct mlx5e_priv *priv,
 
 	(*nhe)->priv = priv;
 	memcpy(&(*nhe)->m_neigh, &e->m_neigh, sizeof(e->m_neigh));
-	INIT_WORK(&(*nhe)->neigh_update_work, mlx5e_rep_neigh_update);
 	spin_lock_init(&(*nhe)->encap_list_lock);
 	INIT_LIST_HEAD(&(*nhe)->encap_list);
 	refcount_set(&(*nhe)->refcnt, 1);

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

@@ -246,8 +246,10 @@ mlx5_tc_ct_rule_to_tuple_nat(struct mlx5_ct_tuple *tuple,
 		case FLOW_ACT_MANGLE_HDR_TYPE_IP6:
 			ip6_offset = (offset - offsetof(struct ipv6hdr, saddr));
 			ip6_offset /= 4;
-			if (ip6_offset < 8)
+			if (ip6_offset < 4)
 				tuple->ip.src_v6.s6_addr32[ip6_offset] = cpu_to_be32(val);
+			else if (ip6_offset < 8)
+				tuple->ip.dst_v6.s6_addr32[ip6_offset - 4] = cpu_to_be32(val);
 			else
 				return -EOPNOTSUPP;
 			break;

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

@@ -217,6 +217,9 @@ static int __mlx5e_add_vlan_rule(struct mlx5e_priv *priv,
 		break;
 	}
 
+	if (WARN_ONCE(*rule_p, "VLAN rule already exists type %d", rule_type))
+		return 0;
+
 	*rule_p = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
 
 	if (IS_ERR(*rule_p)) {
@@ -397,8 +400,7 @@ static void mlx5e_add_vlan_rules(struct mlx5e_priv *priv)
 	for_each_set_bit(i, priv->fs.vlan.active_svlans, VLAN_N_VID)
 		mlx5e_add_vlan_rule(priv, MLX5E_VLAN_RULE_TYPE_MATCH_STAG_VID, i);
 
-	if (priv->fs.vlan.cvlan_filter_disabled &&
-	    !(priv->netdev->flags & IFF_PROMISC))
+	if (priv->fs.vlan.cvlan_filter_disabled)
 		mlx5e_add_any_vid_rules(priv);
 }
 
@@ -415,8 +417,12 @@ static void mlx5e_del_vlan_rules(struct mlx5e_priv *priv)
 	for_each_set_bit(i, priv->fs.vlan.active_svlans, VLAN_N_VID)
 		mlx5e_del_vlan_rule(priv, MLX5E_VLAN_RULE_TYPE_MATCH_STAG_VID, i);
 
-	if (priv->fs.vlan.cvlan_filter_disabled &&
-	    !(priv->netdev->flags & IFF_PROMISC))
+	WARN_ON_ONCE(!(test_bit(MLX5E_STATE_DESTROYING, &priv->state)));
+
+	/* must be called after DESTROY bit is set and
+	 * set_rx_mode is called and flushed
+	 */
+	if (priv->fs.vlan.cvlan_filter_disabled)
 		mlx5e_del_any_vid_rules(priv);
 }
 

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

@@ -246,12 +246,17 @@ static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq,
 
 static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
 				 u64 npages, u8 page_shift,
-				 struct mlx5_core_mkey *umr_mkey)
+				 struct mlx5_core_mkey *umr_mkey,
+				 dma_addr_t filler_addr)
 {
-	int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
+	struct mlx5_mtt *mtt;
+	int inlen;
 	void *mkc;
 	u32 *in;
 	int err;
+	int i;
+
+	inlen = MLX5_ST_SZ_BYTES(create_mkey_in) + sizeof(*mtt) * npages;
 
 	in = kvzalloc(inlen, GFP_KERNEL);
 	if (!in)
@@ -271,6 +276,18 @@ static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
 	MLX5_SET(mkc, mkc, translations_octword_size,
 		 MLX5_MTT_OCTW(npages));
 	MLX5_SET(mkc, mkc, log_page_size, page_shift);
+	MLX5_SET(create_mkey_in, in, translations_octword_actual_size,
+		 MLX5_MTT_OCTW(npages));
+
+	/* Initialize the mkey with all MTTs pointing to a default
+	 * page (filler_addr). When the channels are activated, UMR
+	 * WQEs will redirect the RX WQEs to the actual memory from
+	 * the RQ's pool, while the gaps (wqe_overflow) remain mapped
+	 * to the default page.
+	 */
+	mtt = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
+	for (i = 0 ; i < npages ; i++)
+		mtt[i].ptag = cpu_to_be64(filler_addr);
 
 	err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
 
@@ -282,7 +299,8 @@ static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq
 {
 	u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->mpwqe.wq));
 
-	return mlx5e_create_umr_mkey(mdev, num_mtts, PAGE_SHIFT, &rq->umr_mkey);
+	return mlx5e_create_umr_mkey(mdev, num_mtts, PAGE_SHIFT, &rq->umr_mkey,
+				     rq->wqe_overflow.addr);
 }
 
 static inline u64 mlx5e_get_mpwqe_offset(struct mlx5e_rq *rq, u16 wqe_ix)
@@ -350,6 +368,28 @@ static void mlx5e_rq_err_cqe_work(struct work_struct *recover_work)
 	mlx5e_reporter_rq_cqe_err(rq);
 }
 
+static int mlx5e_alloc_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
+{
+	rq->wqe_overflow.page = alloc_page(GFP_KERNEL);
+	if (!rq->wqe_overflow.page)
+		return -ENOMEM;
+
+	rq->wqe_overflow.addr = dma_map_page(rq->pdev, rq->wqe_overflow.page, 0,
+					     PAGE_SIZE, rq->buff.map_dir);
+	if (dma_mapping_error(rq->pdev, rq->wqe_overflow.addr)) {
+		__free_page(rq->wqe_overflow.page);
+		return -ENOMEM;
+	}
+	return 0;
+}
+
+static void mlx5e_free_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
+{
+	 dma_unmap_page(rq->pdev, rq->wqe_overflow.addr, PAGE_SIZE,
+			rq->buff.map_dir);
+	 __free_page(rq->wqe_overflow.page);
+}
+
 static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 			  struct mlx5e_params *params,
 			  struct mlx5e_xsk_param *xsk,
@@ -396,7 +436,7 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		rq_xdp_ix += params->num_channels * MLX5E_RQ_GROUP_XSK;
 	err = xdp_rxq_info_reg(&rq->xdp_rxq, rq->netdev, rq_xdp_ix);
 	if (err < 0)
-		goto err_rq_wq_destroy;
+		goto err_rq_xdp_prog;
 
 	rq->buff.map_dir = params->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
 	rq->buff.headroom = mlx5e_get_rq_headroom(mdev, params, xsk);
@@ -406,6 +446,10 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
 		err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->mpwqe.wq,
 					&rq->wq_ctrl);
+		if (err)
+			goto err_rq_xdp;
+
+		err = mlx5e_alloc_mpwqe_rq_drop_page(rq);
 		if (err)
 			goto err_rq_wq_destroy;
 
@@ -424,18 +468,18 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 		err = mlx5e_create_rq_umr_mkey(mdev, rq);
 		if (err)
-			goto err_rq_wq_destroy;
+			goto err_rq_drop_page;
 		rq->mkey_be = cpu_to_be32(rq->umr_mkey.key);
 
 		err = mlx5e_rq_alloc_mpwqe_info(rq, c);
 		if (err)
-			goto err_free;
+			goto err_rq_mkey;
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
 		err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
 					 &rq->wq_ctrl);
 		if (err)
-			goto err_rq_wq_destroy;
+			goto err_rq_xdp;
 
 		rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
 
@@ -450,19 +494,19 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 				      GFP_KERNEL, cpu_to_node(c->cpu));
 		if (!rq->wqe.frags) {
 			err = -ENOMEM;
-			goto err_free;
+			goto err_rq_wq_destroy;
 		}
 
 		err = mlx5e_init_di_list(rq, wq_sz, c->cpu);
 		if (err)
-			goto err_free;
+			goto err_rq_frags;
 
 		rq->mkey_be = c->mkey_be;
 	}
 
 	err = mlx5e_rq_set_handlers(rq, params, xsk);
 	if (err)
-		goto err_free;
+		goto err_free_by_rq_type;
 
 	if (xsk) {
 		err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
@@ -486,13 +530,13 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		if (IS_ERR(rq->page_pool)) {
 			err = PTR_ERR(rq->page_pool);
 			rq->page_pool = NULL;
-			goto err_free;
+			goto err_free_by_rq_type;
 		}
 		err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
 						 MEM_TYPE_PAGE_POOL, rq->page_pool);
 	}
 	if (err)
-		goto err_free;
+		goto err_free_by_rq_type;
 
 	for (i = 0; i < wq_sz; i++) {
 		if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
@@ -542,23 +586,27 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	return 0;
 
-err_free:
+err_free_by_rq_type:
 	switch (rq->wq_type) {
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
 		kvfree(rq->mpwqe.info);
+err_rq_mkey:
 		mlx5_core_destroy_mkey(mdev, &rq->umr_mkey);
+err_rq_drop_page:
+		mlx5e_free_mpwqe_rq_drop_page(rq);
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
-		kvfree(rq->wqe.frags);
 		mlx5e_free_di_list(rq);
+err_rq_frags:
+		kvfree(rq->wqe.frags);
 	}
-
 err_rq_wq_destroy:
+	mlx5_wq_destroy(&rq->wq_ctrl);
+err_rq_xdp:
+	xdp_rxq_info_unreg(&rq->xdp_rxq);
+err_rq_xdp_prog:
 	if (params->xdp_prog)
 		bpf_prog_put(params->xdp_prog);
-	xdp_rxq_info_unreg(&rq->xdp_rxq);
-	page_pool_destroy(rq->page_pool);
-	mlx5_wq_destroy(&rq->wq_ctrl);
 
 	return err;
 }
@@ -580,6 +628,7 @@ static void mlx5e_free_rq(struct mlx5e_rq *rq)
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
 		kvfree(rq->mpwqe.info);
 		mlx5_core_destroy_mkey(rq->mdev, &rq->umr_mkey);
+		mlx5e_free_mpwqe_rq_drop_page(rq);
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
 		kvfree(rq->wqe.frags);
@@ -4177,6 +4226,21 @@ int mlx5e_get_vf_stats(struct net_device *dev,
 }
 #endif
 
+static bool mlx5e_gre_tunnel_inner_proto_offload_supported(struct mlx5_core_dev *mdev,
+							   struct sk_buff *skb)
+{
+	switch (skb->inner_protocol) {
+	case htons(ETH_P_IP):
+	case htons(ETH_P_IPV6):
+	case htons(ETH_P_TEB):
+		return true;
+	case htons(ETH_P_MPLS_UC):
+	case htons(ETH_P_MPLS_MC):
+		return MLX5_CAP_ETH(mdev, tunnel_stateless_mpls_over_gre);
+	}
+	return false;
+}
+
 static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
 						     struct sk_buff *skb,
 						     netdev_features_t features)
@@ -4199,7 +4263,9 @@ static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
 
 	switch (proto) {
 	case IPPROTO_GRE:
-		return features;
+		if (mlx5e_gre_tunnel_inner_proto_offload_supported(priv->mdev, skb))
+			return features;
+		break;
 	case IPPROTO_IPIP:
 	case IPPROTO_IPV6:
 		if (mlx5e_tunnel_proto_supported(priv->mdev, IPPROTO_IPIP))

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

@@ -135,12 +135,6 @@ struct mlx5e_neigh_hash_entry {
 	/* encap list sharing the same neigh */
 	struct list_head encap_list;
 
-	/* valid only when the neigh reference is taken during
-	 * neigh_update_work workqueue callback.
-	 */
-	struct neighbour *n;
-	struct work_struct neigh_update_work;
-
 	/* neigh hash entry can be deleted only when the refcount is zero.
 	 * refcount is needed to avoid neigh hash entry removal by TC, while
 	 * it's used by the neigh notification call.

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

@@ -189,6 +189,29 @@ u32 mlx5_eq_poll_irq_disabled(struct mlx5_eq_comp *eq)
 	return count_eqe;
 }
 
+static void mlx5_eq_async_int_lock(struct mlx5_eq_async *eq, unsigned long *flags)
+	__acquires(&eq->lock)
+{
+	if (in_irq())
+		spin_lock(&eq->lock);
+	else
+		spin_lock_irqsave(&eq->lock, *flags);
+}
+
+static void mlx5_eq_async_int_unlock(struct mlx5_eq_async *eq, unsigned long *flags)
+	__releases(&eq->lock)
+{
+	if (in_irq())
+		spin_unlock(&eq->lock);
+	else
+		spin_unlock_irqrestore(&eq->lock, *flags);
+}
+
+enum async_eq_nb_action {
+	ASYNC_EQ_IRQ_HANDLER = 0,
+	ASYNC_EQ_RECOVER = 1,
+};
+
 static int mlx5_eq_async_int(struct notifier_block *nb,
 			     unsigned long action, void *data)
 {
@@ -198,11 +221,14 @@ static int mlx5_eq_async_int(struct notifier_block *nb,
 	struct mlx5_eq_table *eqt;
 	struct mlx5_core_dev *dev;
 	struct mlx5_eqe *eqe;
+	unsigned long flags;
 	int num_eqes = 0;
 
 	dev = eq->dev;
 	eqt = dev->priv.eq_table;
 
+	mlx5_eq_async_int_lock(eq_async, &flags);
+
 	eqe = next_eqe_sw(eq);
 	if (!eqe)
 		goto out;
@@ -223,8 +249,19 @@ static int mlx5_eq_async_int(struct notifier_block *nb,
 
 out:
 	eq_update_ci(eq, 1);
+	mlx5_eq_async_int_unlock(eq_async, &flags);
 
-	return 0;
+	return unlikely(action == ASYNC_EQ_RECOVER) ? num_eqes : 0;
+}
+
+void mlx5_cmd_eq_recover(struct mlx5_core_dev *dev)
+{
+	struct mlx5_eq_async *eq = &dev->priv.eq_table->cmd_eq;
+	int eqes;
+
+	eqes = mlx5_eq_async_int(&eq->irq_nb, ASYNC_EQ_RECOVER, NULL);
+	if (eqes)
+		mlx5_core_warn(dev, "Recovered %d EQEs on cmd_eq\n", eqes);
 }
 
 static void init_eq_buf(struct mlx5_eq *eq)
@@ -569,6 +606,7 @@ setup_async_eq(struct mlx5_core_dev *dev, struct mlx5_eq_async *eq,
 	int err;
 
 	eq->irq_nb.notifier_call = mlx5_eq_async_int;
+	spin_lock_init(&eq->lock);
 
 	err = create_async_eq(dev, &eq->core, param);
 	if (err) {
@@ -656,8 +694,10 @@ static void destroy_async_eqs(struct mlx5_core_dev *dev)
 
 	cleanup_async_eq(dev, &table->pages_eq, "pages");
 	cleanup_async_eq(dev, &table->async_eq, "async");
+	mlx5_cmd_allowed_opcode(dev, MLX5_CMD_OP_DESTROY_EQ);
 	mlx5_cmd_use_polling(dev);
 	cleanup_async_eq(dev, &table->cmd_eq, "cmd");
+	mlx5_cmd_allowed_opcode(dev, CMD_ALLOWED_OPCODE_ALL);
 	mlx5_eq_notifier_unregister(dev, &table->cq_err_nb);
 }
 

drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h

@@ -37,6 +37,7 @@ struct mlx5_eq {
 struct mlx5_eq_async {
 	struct mlx5_eq          core;
 	struct notifier_block   irq_nb;
+	spinlock_t              lock; /* To avoid irq EQ handle races with resiliency flows */
 };
 
 struct mlx5_eq_comp {
@@ -81,6 +82,7 @@ void mlx5_cq_tasklet_cb(unsigned long data);
 struct cpumask *mlx5_eq_comp_cpumask(struct mlx5_core_dev *dev, int ix);
 
 u32 mlx5_eq_poll_irq_disabled(struct mlx5_eq_comp *eq);
+void mlx5_cmd_eq_recover(struct mlx5_core_dev *dev);
 void mlx5_eq_synchronize_async_irq(struct mlx5_core_dev *dev);
 void mlx5_eq_synchronize_cmd_irq(struct mlx5_core_dev *dev);
 

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

@@ -432,7 +432,7 @@ static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
 	u32 npages;
 	u32 i = 0;
 
-	if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
+	if (!mlx5_cmd_is_down(dev))
 		return mlx5_cmd_exec(dev, in, in_size, out, out_size);
 
 	/* No hard feelings, we want our pages back! */

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

@@ -115,7 +115,7 @@ static int request_irqs(struct mlx5_core_dev *dev, int nvec)
 	return 0;
 
 err_request_irq:
-	for (; i >= 0; i--) {
+	while (i--) {
 		struct mlx5_irq *irq = mlx5_irq_get(dev, i);
 		int irqn = pci_irq_vector(dev->pdev, i);
 

include/linux/mlx5/driver.h

@@ -767,6 +767,8 @@ struct mlx5_cmd_work_ent {
 	u64			ts2;
 	u16			op;
 	bool			polling;
+	/* Track the max comp handlers */
+	refcount_t              refcnt;
 };
 
 struct mlx5_pas {
@@ -933,6 +935,7 @@ int mlx5_cmd_exec(struct mlx5_core_dev *dev, void *in, int in_size, void *out,
 int mlx5_cmd_exec_polling(struct mlx5_core_dev *dev, void *in, int in_size,
 			  void *out, int out_size);
 void mlx5_cmd_mbox_status(void *out, u8 *status, u32 *syndrome);
+bool mlx5_cmd_is_down(struct mlx5_core_dev *dev);
 
 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type);
 int mlx5_cmd_alloc_uar(struct mlx5_core_dev *dev, u32 *uarn);