mlx5: Numa domain improvements.

Properly allocate all mlx5en(4) structures from correct numa domain.

While at it cleanup unused numa domain integers deriving from the
Linux version of mlx5en(4).

Reviewed by:	kib
Sponsored by:	Mellanox Technologies // NVIDIA Networking

(cherry picked from commit 7c3eff94bda8bb746bfa7a5edc81b014e2dc97f6)

7 files changed, 43 insertions, 57 deletions

diff --git a/sys/dev/mlx5/driver.h b/sys/dev/mlx5/driver.h
index d21a4aa075c0..614a308e8e2f 100644
--- a/sys/dev/mlx5/driver.h
+++ b/sys/dev/mlx5/driver.h
@@ -985,8 +985,6 @@ void mlx5_drain_health_recovery(struct mlx5_core_dev *dev);
 void mlx5_trigger_health_work(struct mlx5_core_dev *dev);
 void mlx5_trigger_health_watchdog(struct mlx5_core_dev *dev);
 
-#define	mlx5_buf_alloc_node(dev, size, direct, buf, node) \
-	mlx5_buf_alloc(dev, size, direct, buf)
 int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, int max_direct,
 		   struct mlx5_buf *buf);
 void mlx5_buf_free(struct mlx5_core_dev *dev, struct mlx5_buf *buf);
@@ -1072,10 +1070,14 @@ void mlx5_eq_debugfs_cleanup(struct mlx5_core_dev *dev);
 int mlx5_cq_debugfs_init(struct mlx5_core_dev *dev);
 void mlx5_cq_debugfs_cleanup(struct mlx5_core_dev *dev);
 int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db);
-int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db,
-		       int node);
 void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db);
 
+static inline struct domainset *
+mlx5_dev_domainset(struct mlx5_core_dev *mdev)
+{
+	return (linux_get_vm_domain_set(mdev->priv.numa_node));
+}
+
 const char *mlx5_command_str(int command);
 int mlx5_cmdif_debugfs_init(struct mlx5_core_dev *dev);
 void mlx5_cmdif_debugfs_cleanup(struct mlx5_core_dev *dev);
diff --git a/sys/dev/mlx5/mlx5_core/mlx5_alloc.c b/sys/dev/mlx5/mlx5_core/mlx5_alloc.c
index 1dabbd9f586f..99e39b80a4ec 100644
--- a/sys/dev/mlx5/mlx5_core/mlx5_alloc.c
+++ b/sys/dev/mlx5/mlx5_core/mlx5_alloc.c
@@ -147,8 +147,7 @@ void mlx5_buf_free(struct mlx5_core_dev *dev, struct mlx5_buf *buf)
 }
 EXPORT_SYMBOL_GPL(mlx5_buf_free);
 
-static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
-						 int node)
+static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev)
 {
 	struct mlx5_db_pgdir *pgdir;
 
@@ -199,7 +198,7 @@ static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir,
 	return 0;
 }
 
-int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node)
+int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db)
 {
 	struct mlx5_db_pgdir *pgdir;
 	int ret = 0;
@@ -210,7 +209,7 @@ int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node)
 		if (!mlx5_alloc_db_from_pgdir(pgdir, db))
 			goto out;
 
-	pgdir = mlx5_alloc_db_pgdir(dev, node);
+	pgdir = mlx5_alloc_db_pgdir(dev);
 	if (!pgdir) {
 		ret = -ENOMEM;
 		goto out;
@@ -226,12 +225,6 @@ out:
 
 	return ret;
 }
-EXPORT_SYMBOL_GPL(mlx5_db_alloc_node);
-
-int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db)
-{
-	return mlx5_db_alloc_node(dev, db, dev->priv.numa_node);
-}
 EXPORT_SYMBOL_GPL(mlx5_db_alloc);
 
 void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db)
diff --git a/sys/dev/mlx5/mlx5_core/mlx5_main.c b/sys/dev/mlx5/mlx5_core/mlx5_main.c
index 755e8844a892..7527af524456 100644
--- a/sys/dev/mlx5/mlx5_core/mlx5_main.c
+++ b/sys/dev/mlx5/mlx5_core/mlx5_main.c
@@ -82,8 +82,6 @@ SYSCTL_INT(_hw_mlx5, OID_AUTO, fast_unload_enabled, CTLFLAG_RWTUN,
     &mlx5_fast_unload_enabled, 0,
     "Set to enable fast unload. Clear to disable.");
 
-#define NUMA_NO_NODE       -1
-
 static LIST_HEAD(intf_list);
 static LIST_HEAD(dev_list);
 static DEFINE_MUTEX(intf_mutex);
@@ -653,7 +651,7 @@ static int alloc_comp_eqs(struct mlx5_core_dev *dev)
 	ncomp_vec = table->num_comp_vectors;
 	nent = MLX5_COMP_EQ_SIZE;
 	for (i = 0; i < ncomp_vec; i++) {
-		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
+		eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, dev->priv.numa_node);
 
 		err = mlx5_create_map_eq(dev, eq,
 					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0);
@@ -715,7 +713,7 @@ static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
 	struct mlx5_device_context *dev_ctx;
 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
 
-	dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL);
+	dev_ctx = kzalloc_node(sizeof(*dev_ctx), GFP_KERNEL, priv->numa_node);
 	if (!dev_ctx)
 		return;
 
@@ -869,8 +867,6 @@ static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
 	INIT_LIST_HEAD(&priv->pgdir_list);
 	spin_lock_init(&priv->mkey_lock);
 
-	priv->numa_node = NUMA_NO_NODE;
-
 	err = mlx5_pci_enable_device(dev);
 	if (err) {
 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
@@ -1314,14 +1310,20 @@ static int init_one(struct pci_dev *pdev,
 	int num_vfs, sriov_pos;
 #endif
 	int i,err;
+	int numa_node;
 	struct sysctl_oid *pme_sysctl_node;
 	struct sysctl_oid *pme_err_sysctl_node;
 	struct sysctl_oid *cap_sysctl_node;
 	struct sysctl_oid *current_cap_sysctl_node;
 	struct sysctl_oid *max_cap_sysctl_node;
 
-	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
+	numa_node = dev_to_node(&pdev->dev);
+
+	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, numa_node);
+
 	priv = &dev->priv;
+	priv->numa_node = numa_node;
+
 	if (id)
 		priv->pci_dev_data = id->driver_data;
 
diff --git a/sys/dev/mlx5/mlx5_core/mlx5_wq.c b/sys/dev/mlx5/mlx5_core/mlx5_wq.c
index 6cbb7c0000ae..131ed9bfce2a 100644
--- a/sys/dev/mlx5/mlx5_core/mlx5_wq.c
+++ b/sys/dev/mlx5/mlx5_core/mlx5_wq.c
@@ -69,15 +69,14 @@ int mlx5_wq_cyc_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param,
 	wq->log_stride = MLX5_GET(wq, wqc, log_wq_stride);
 	wq->sz_m1 = (1 << MLX5_GET(wq, wqc, log_wq_sz)) - 1;
 
-	err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node);
+	err = mlx5_db_alloc(mdev, &wq_ctrl->db);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err);
 		return err;
 	}
 
-	err = mlx5_buf_alloc_node(mdev, mlx5_wq_cyc_get_byte_size(wq),
-				  max_direct, &wq_ctrl->buf,
-				  param->buf_numa_node);
+	err = mlx5_buf_alloc(mdev, mlx5_wq_cyc_get_byte_size(wq),
+			     max_direct, &wq_ctrl->buf);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err);
 		goto err_db_free;
@@ -107,15 +106,14 @@ int mlx5_cqwq_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param,
 	wq->log_sz = MLX5_GET(cqc, cqc, log_cq_size);
 	wq->sz_m1 = (1 << wq->log_sz) - 1;
 
-	err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node);
+	err = mlx5_db_alloc(mdev, &wq_ctrl->db);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err);
 		return err;
 	}
 
-	err = mlx5_buf_alloc_node(mdev, mlx5_cqwq_get_byte_size(wq),
-				  max_direct, &wq_ctrl->buf,
-				  param->buf_numa_node);
+	err = mlx5_buf_alloc(mdev, mlx5_cqwq_get_byte_size(wq),
+			     max_direct, &wq_ctrl->buf);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err);
 		goto err_db_free;
@@ -146,15 +144,14 @@ int mlx5_wq_ll_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param,
 	wq->log_stride = MLX5_GET(wq, wqc, log_wq_stride);
 	wq->sz_m1 = (1 << MLX5_GET(wq, wqc, log_wq_sz)) - 1;
 
-	err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node);
+	err = mlx5_db_alloc(mdev, &wq_ctrl->db);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err);
 		return err;
 	}
 
-	err = mlx5_buf_alloc_node(mdev, mlx5_wq_ll_get_byte_size(wq),
-				  max_direct, &wq_ctrl->buf,
-				  param->buf_numa_node);
+	err = mlx5_buf_alloc(mdev, mlx5_wq_ll_get_byte_size(wq),
+			     max_direct, &wq_ctrl->buf);
 	if (err) {
 		mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err);
 		goto err_db_free;
diff --git a/sys/dev/mlx5/mlx5_core/wq.h b/sys/dev/mlx5/mlx5_core/wq.h
index 13f26b0dd9a1..f513e4d7f6b4 100644
--- a/sys/dev/mlx5/mlx5_core/wq.h
+++ b/sys/dev/mlx5/mlx5_core/wq.h
@@ -32,8 +32,6 @@
 
 struct mlx5_wq_param {
 	int		linear;
-	int		buf_numa_node;
-	int		db_numa_node;
 };
 
 struct mlx5_wq_ctrl {
diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
index 9345798006f2..170fb03be968 100644
--- a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
+++ b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
@@ -1261,7 +1261,8 @@ mlx5e_create_rq(struct mlx5e_channel *c,
 	if (err)
 		goto err_rq_wq_destroy;
 
-	rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
+	rq->mbuf = malloc_domainset(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN,
+	    mlx5_dev_domainset(mdev), M_WAITOK | M_ZERO);
 	for (i = 0; i != wq_sz; i++) {
 		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
 		int j;
@@ -1525,7 +1526,8 @@ mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
 	int err;
 	int x;
 
-	sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
+	sq->mbuf = malloc_domainset(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN,
+	    mlx5_dev_domainset(sq->priv->mdev), M_WAITOK | M_ZERO);
 
 	/* Create DMA descriptor MAPs */
 	for (x = 0; x != wq_sz; x++) {
@@ -1619,6 +1621,11 @@ mlx5e_create_sq(struct mlx5e_channel *c,
 	    &sq->dma_tag)))
 		goto done;
 
+	sq->mkey_be = cpu_to_be32(priv->mr.key);
+	sq->ifp = priv->ifp;
+	sq->priv = priv;
+	sq->tc = tc;
+
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, &sq->wq,
 	    &sq->wq_ctrl);
 	if (err)
@@ -1630,11 +1637,6 @@ mlx5e_create_sq(struct mlx5e_channel *c,
 	if (err)
 		goto err_sq_wq_destroy;
 
-	sq->mkey_be = cpu_to_be32(priv->mr.key);
-	sq->ifp = priv->ifp;
-	sq->priv = priv;
-	sq->tc = tc;
-
 	mlx5e_update_sq_inline(sq);
 
 	snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
@@ -1927,9 +1929,6 @@ mlx5e_create_cq(struct mlx5e_priv *priv,
 	int err;
 	u32 i;
 
-	param->wq.buf_numa_node = 0;
-	param->wq.db_numa_node = 0;
-
 	err = mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
 	if (err)
 		return (err);
@@ -2292,8 +2291,6 @@ mlx5e_build_rq_param(struct mlx5e_priv *priv,
 	MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
 	MLX5_SET(wq, wq, pd, priv->pdn);
 
-	param->wq.buf_numa_node = 0;
-	param->wq.db_numa_node = 0;
 	param->wq.linear = 1;
 }
 
@@ -2308,8 +2305,6 @@ mlx5e_build_sq_param(struct mlx5e_priv *priv,
 	MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
 	MLX5_SET(wq, wq, pd, priv->pdn);
 
-	param->wq.buf_numa_node = 0;
-	param->wq.db_numa_node = 0;
 	param->wq.linear = 1;
 }
 
@@ -4430,13 +4425,14 @@ mlx5e_create_ifp(struct mlx5_core_dev *mdev)
 		mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
 		return (NULL);
 	}
+
 	/*
 	 * Try to allocate the priv and make room for worst-case
 	 * number of channel structures:
 	 */
-	priv = malloc(sizeof(*priv) +
+	priv = malloc_domainset(sizeof(*priv) +
 	    (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
-	    M_MLX5EN, M_WAITOK | M_ZERO);
+	    M_MLX5EN, mlx5_dev_domainset(mdev), M_WAITOK | M_ZERO);
 
 	ifp = priv->ifp = if_alloc_dev(IFT_ETHER, mdev->pdev->dev.bsddev);
 	if (ifp == NULL) {
diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c b/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
index bd66b8ecd725..fa7b812fbdf1 100644
--- a/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
+++ b/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
@@ -51,8 +51,6 @@ mlx5e_rl_build_sq_param(struct mlx5e_rl_priv_data *rl,
 	MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
 	MLX5_SET(wq, wq, pd, rl->priv->pdn);
 
-	param->wq.buf_numa_node = 0;
-	param->wq.db_numa_node = 0;
 	param->wq.linear = 1;
 }
 
@@ -116,6 +114,10 @@ mlx5e_rl_create_sq(struct mlx5e_priv *priv, struct mlx5e_sq *sq,
 	    &sq->dma_tag)))
 		goto done;
 
+	sq->mkey_be = cpu_to_be32(priv->mr.key);
+	sq->ifp = priv->ifp;
+	sq->priv = priv;
+
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, &sq->wq,
 	    &sq->wq_ctrl);
 	if (err)
@@ -127,10 +129,6 @@ mlx5e_rl_create_sq(struct mlx5e_priv *priv, struct mlx5e_sq *sq,
 	if (err)
 		goto err_sq_wq_destroy;
 
-	sq->mkey_be = cpu_to_be32(priv->mr.key);
-	sq->ifp = priv->ifp;
-	sq->priv = priv;
-
 	mlx5e_update_sq_inline(sq);
 
 	return (0);