mlx4: Structures and init/teardown for VF resource quotas

This is step #1 for implementing SRIOV resource quotas for VFs.

Quotas are implemented per resource type for VFs and the PF, to prevent
any entity from simply grabbing all the resources for itself and leaving
the other entities unable to obtain such resources.

Resources which are allocated using quotas:  QPs, CQs, SRQs, MPTs, MTTs, MAC,
                                             VLAN, and Counters.

The quota system works as follows:
Each entity (VF or PF) is given a max number of a given resource (its quota),
and a guaranteed minimum number for each resource (starvation prevention).

For QPs, CQs, SRQs, MPTs and MTTs:
50% of the available quantity for the resource is divided equally among
the PF and all the active VFs (i.e., the number of VFs in the mlx4_core module
parameter "num_vfs"). This 50% represents the "guaranteed minimum" pool.
The other 50% is the "free pool", allocated on a first-come-first-serve basis.
For each VF/PF, resources are first allocated from its "guaranteed-minimum"
pool. When that pool is exhausted, the driver attempts to allocate from
the resource "free-pool".

The quota (i.e., max) for the VFs and the PF is:
  The free-pool amount (50% of the real max) + the guaranteed minimum

For MACs:
  Guarantee 2 MACs per VF/PF per port. As a result, since we have only
  128 MACs per port, reduce the allowable number of VFs from 64 to 63.
  Any remaining MACs are put into a free pool.

For VLANs:
  For the PF, the per-port quota is 128 and guarantee is 64
     (to allow the PF to register at least a VLAN per VF in VST mode).
  For the VFs, the per-port quota is 64 and the guarantee is 0.
      We assume that VGT VFs are trusted not to abuse the VLAN resource.

For Counters:
  For all functions (PF and VFs), the quota is 128 and the guarantee is 0.

In this patch, we define the needed structures, which are added to the
resource-tracker struct.  In addition, we do initialization
for the resource quota, and adjust the query_device response to use quotas
rather than resource maxima.

As part of the implementation, we introduce a new field in
mlx4_dev: quotas.  This field holds the resource quotas used
to report maxima to the upper layers (ib_core, via query_device).

The HCA maxima of these values are passed to the VFs (via
QUERY_HCA) so that they may continue to use these in handling
QPs, CQs, SRQs and MPTs.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/infiniband/hw/mlx4/main.c

@@ -177,18 +177,18 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 
 	props->max_mr_size	   = ~0ull;
 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
-	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
+	props->max_qp		   = dev->dev->quotas.qp;
 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
 					 dev->dev->caps.max_rq_sg);
-	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
+	props->max_cq		   = dev->dev->quotas.cq;
 	props->max_cqe		   = dev->dev->caps.max_cqes;
-	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
+	props->max_mr		   = dev->dev->quotas.mpt;
 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
-	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
+	props->max_srq		   = dev->dev->quotas.srq;
 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;

drivers/net/ethernet/mellanox/mlx4/fw.c

@@ -177,6 +177,7 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 				struct mlx4_cmd_mailbox *outbox,
 				struct mlx4_cmd_info *cmd)
 {
+	struct mlx4_priv *priv = mlx4_priv(dev);
 	u8	field;
 	u32	size;
 	int	err = 0;
@@ -250,13 +251,13 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 		field = 0; /* protected FMR support not available as yet */
 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FMR_OFFSET);
 
-		size = dev->caps.num_qps;
+		size = priv->mfunc.master.res_tracker.res_alloc[RES_QP].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
 
-		size = dev->caps.num_srqs;
+		size = priv->mfunc.master.res_tracker.res_alloc[RES_SRQ].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
 
-		size = dev->caps.num_cqs;
+		size = priv->mfunc.master.res_tracker.res_alloc[RES_CQ].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
 
 		size = dev->caps.num_eqs;
@@ -265,10 +266,10 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 		size = dev->caps.reserved_eqs;
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
 
-		size = dev->caps.num_mpts;
+		size = priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
 
-		size = dev->caps.num_mtts;
+		size = priv->mfunc.master.res_tracker.res_alloc[RES_MTT].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
 
 		size = dev->caps.num_mgms + dev->caps.num_amgms;

drivers/net/ethernet/mellanox/mlx4/main.c

@@ -562,13 +562,17 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
 	}
 
 	dev->caps.num_ports		= func_cap.num_ports;
-	dev->caps.num_qps		= func_cap.qp_quota;
-	dev->caps.num_srqs		= func_cap.srq_quota;
-	dev->caps.num_cqs		= func_cap.cq_quota;
-	dev->caps.num_eqs               = func_cap.max_eq;
-	dev->caps.reserved_eqs          = func_cap.reserved_eq;
-	dev->caps.num_mpts		= func_cap.mpt_quota;
-	dev->caps.num_mtts		= func_cap.mtt_quota;
+	dev->quotas.qp			= func_cap.qp_quota;
+	dev->quotas.srq			= func_cap.srq_quota;
+	dev->quotas.cq			= func_cap.cq_quota;
+	dev->quotas.mpt			= func_cap.mpt_quota;
+	dev->quotas.mtt			= func_cap.mtt_quota;
+	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
+	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
+	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
+	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
+	dev->caps.num_eqs		= func_cap.max_eq;
+	dev->caps.reserved_eqs		= func_cap.reserved_eq;
 	dev->caps.num_pds               = MLX4_NUM_PDS;
 	dev->caps.num_mgms              = 0;
 	dev->caps.num_amgms             = 0;
@@ -2102,9 +2106,15 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
 			"aborting.\n");
 		return err;
 	}
-	if (num_vfs > MLX4_MAX_NUM_VF) {
-		printk(KERN_ERR "There are more VF's (%d) than allowed(%d)\n",
-		       num_vfs, MLX4_MAX_NUM_VF);
+
+	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
+	 * per port, we must limit the number of VFs to 63 (since their are
+	 * 128 MACs)
+	 */
+	if (num_vfs >= MLX4_MAX_NUM_VF) {
+		dev_err(&pdev->dev,
+			"Requested more VF's (%d) than allowed (%d)\n",
+			num_vfs, MLX4_MAX_NUM_VF - 1);
 		return -EINVAL;
 	}
 
@@ -2322,6 +2332,8 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
 	if (err)
 		goto err_steer;
 
+	mlx4_init_quotas(dev);
+
 	for (port = 1; port <= dev->caps.num_ports; port++) {
 		err = mlx4_init_port_info(dev, port);
 		if (err)

drivers/net/ethernet/mellanox/mlx4/mlx4.h

@@ -504,12 +504,27 @@ struct slave_list {
 	struct list_head res_list[MLX4_NUM_OF_RESOURCE_TYPE];
 };
 
+struct resource_allocator {
+	union {
+		int res_reserved;
+		int res_port_rsvd[MLX4_MAX_PORTS];
+	};
+	union {
+		int res_free;
+		int res_port_free[MLX4_MAX_PORTS];
+	};
+	int *quota;
+	int *allocated;
+	int *guaranteed;
+};
+
 struct mlx4_resource_tracker {
 	spinlock_t lock;
 	/* tree for each resources */
 	struct rb_root res_tree[MLX4_NUM_OF_RESOURCE_TYPE];
 	/* num_of_slave's lists, one per slave */
 	struct slave_list *slave_list;
+	struct resource_allocator res_alloc[MLX4_NUM_OF_RESOURCE_TYPE];
 };
 
 #define SLAVE_EVENT_EQ_SIZE	128
@@ -1253,4 +1268,6 @@ static inline spinlock_t *mlx4_tlock(struct mlx4_dev *dev)
 
 void mlx4_vf_immed_vlan_work_handler(struct work_struct *_work);
 
+void mlx4_init_quotas(struct mlx4_dev *dev);
+
 #endif /* MLX4_H */

drivers/net/ethernet/mellanox/mlx4/qp.c

@@ -480,8 +480,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	*/
 
 	err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps,
-			       (1 << 23) - 1, dev->phys_caps.base_sqpn + 8 +
-			       16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev),
+			       (1 << 23) - 1, mlx4_num_reserved_sqps(dev),
 			       reserved_from_top);
 	if (err)
 		return err;

drivers/net/ethernet/mellanox/mlx4/resource_tracker.c

@@ -284,10 +284,59 @@ static const char *ResourceType(enum mlx4_resource rt)
 }
 
 static void rem_slave_vlans(struct mlx4_dev *dev, int slave);
+static inline void initialize_res_quotas(struct mlx4_dev *dev,
+					 struct resource_allocator *res_alloc,
+					 enum mlx4_resource res_type,
+					 int vf, int num_instances)
+{
+	res_alloc->guaranteed[vf] = num_instances / (2 * (dev->num_vfs + 1));
+	res_alloc->quota[vf] = (num_instances / 2) + res_alloc->guaranteed[vf];
+	if (vf == mlx4_master_func_num(dev)) {
+		res_alloc->res_free = num_instances;
+		if (res_type == RES_MTT) {
+			/* reserved mtts will be taken out of the PF allocation */
+			res_alloc->res_free += dev->caps.reserved_mtts;
+			res_alloc->guaranteed[vf] += dev->caps.reserved_mtts;
+			res_alloc->quota[vf] += dev->caps.reserved_mtts;
+		}
+	}
+}
+
+void mlx4_init_quotas(struct mlx4_dev *dev)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	int pf;
+
+	/* quotas for VFs are initialized in mlx4_slave_cap */
+	if (mlx4_is_slave(dev))
+		return;
+
+	if (!mlx4_is_mfunc(dev)) {
+		dev->quotas.qp = dev->caps.num_qps - dev->caps.reserved_qps -
+			mlx4_num_reserved_sqps(dev);
+		dev->quotas.cq = dev->caps.num_cqs - dev->caps.reserved_cqs;
+		dev->quotas.srq = dev->caps.num_srqs - dev->caps.reserved_srqs;
+		dev->quotas.mtt = dev->caps.num_mtts - dev->caps.reserved_mtts;
+		dev->quotas.mpt = dev->caps.num_mpts - dev->caps.reserved_mrws;
+		return;
+	}
+
+	pf = mlx4_master_func_num(dev);
+	dev->quotas.qp =
+		priv->mfunc.master.res_tracker.res_alloc[RES_QP].quota[pf];
+	dev->quotas.cq =
+		priv->mfunc.master.res_tracker.res_alloc[RES_CQ].quota[pf];
+	dev->quotas.srq =
+		priv->mfunc.master.res_tracker.res_alloc[RES_SRQ].quota[pf];
+	dev->quotas.mtt =
+		priv->mfunc.master.res_tracker.res_alloc[RES_MTT].quota[pf];
+	dev->quotas.mpt =
+		priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[pf];
+}
 int mlx4_init_resource_tracker(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
-	int i;
+	int i, j;
 	int t;
 
 	priv->mfunc.master.res_tracker.slave_list =
@@ -308,8 +357,104 @@ int mlx4_init_resource_tracker(struct mlx4_dev *dev)
 	for (i = 0 ; i < MLX4_NUM_OF_RESOURCE_TYPE; i++)
 		priv->mfunc.master.res_tracker.res_tree[i] = RB_ROOT;
 
+	for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
+		struct resource_allocator *res_alloc =
+			&priv->mfunc.master.res_tracker.res_alloc[i];
+		res_alloc->quota = kmalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
+		res_alloc->guaranteed = kmalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
+		if (i == RES_MAC || i == RES_VLAN)
+			res_alloc->allocated = kzalloc(MLX4_MAX_PORTS *
+						       (dev->num_vfs + 1) * sizeof(int),
+							GFP_KERNEL);
+		else
+			res_alloc->allocated = kzalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
+
+		if (!res_alloc->quota || !res_alloc->guaranteed ||
+		    !res_alloc->allocated)
+			goto no_mem_err;
+
+		for (t = 0; t < dev->num_vfs + 1; t++) {
+			switch (i) {
+			case RES_QP:
+				initialize_res_quotas(dev, res_alloc, RES_QP,
+						      t, dev->caps.num_qps -
+						      dev->caps.reserved_qps -
+						      mlx4_num_reserved_sqps(dev));
+				break;
+			case RES_CQ:
+				initialize_res_quotas(dev, res_alloc, RES_CQ,
+						      t, dev->caps.num_cqs -
+						      dev->caps.reserved_cqs);
+				break;
+			case RES_SRQ:
+				initialize_res_quotas(dev, res_alloc, RES_SRQ,
+						      t, dev->caps.num_srqs -
+						      dev->caps.reserved_srqs);
+				break;
+			case RES_MPT:
+				initialize_res_quotas(dev, res_alloc, RES_MPT,
+						      t, dev->caps.num_mpts -
+						      dev->caps.reserved_mrws);
+				break;
+			case RES_MTT:
+				initialize_res_quotas(dev, res_alloc, RES_MTT,
+						      t, dev->caps.num_mtts -
+						      dev->caps.reserved_mtts);
+				break;
+			case RES_MAC:
+				if (t == mlx4_master_func_num(dev)) {
+					res_alloc->quota[t] = MLX4_MAX_MAC_NUM;
+					res_alloc->guaranteed[t] = 2;
+					for (j = 0; j < MLX4_MAX_PORTS; j++)
+						res_alloc->res_port_free[j] = MLX4_MAX_MAC_NUM;
+				} else {
+					res_alloc->quota[t] = MLX4_MAX_MAC_NUM;
+					res_alloc->guaranteed[t] = 2;
+				}
+				break;
+			case RES_VLAN:
+				if (t == mlx4_master_func_num(dev)) {
+					res_alloc->quota[t] = MLX4_MAX_VLAN_NUM;
+					res_alloc->guaranteed[t] = MLX4_MAX_VLAN_NUM / 2;
+					for (j = 0; j < MLX4_MAX_PORTS; j++)
+						res_alloc->res_port_free[j] =
+							res_alloc->quota[t];
+				} else {
+					res_alloc->quota[t] = MLX4_MAX_VLAN_NUM / 2;
+					res_alloc->guaranteed[t] = 0;
+				}
+				break;
+			case RES_COUNTER:
+				res_alloc->quota[t] = dev->caps.max_counters;
+				res_alloc->guaranteed[t] = 0;
+				if (t == mlx4_master_func_num(dev))
+					res_alloc->res_free = res_alloc->quota[t];
+				break;
+			default:
+				break;
+			}
+			if (i == RES_MAC || i == RES_VLAN) {
+				for (j = 0; j < MLX4_MAX_PORTS; j++)
+					res_alloc->res_port_rsvd[j] +=
+						res_alloc->guaranteed[t];
+			} else {
+				res_alloc->res_reserved += res_alloc->guaranteed[t];
+			}
+		}
+	}
 	spin_lock_init(&priv->mfunc.master.res_tracker.lock);
-	return 0 ;
+	return 0;
+
+no_mem_err:
+	for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
+		kfree(priv->mfunc.master.res_tracker.res_alloc[i].allocated);
+		priv->mfunc.master.res_tracker.res_alloc[i].allocated = NULL;
+		kfree(priv->mfunc.master.res_tracker.res_alloc[i].guaranteed);
+		priv->mfunc.master.res_tracker.res_alloc[i].guaranteed = NULL;
+		kfree(priv->mfunc.master.res_tracker.res_alloc[i].quota);
+		priv->mfunc.master.res_tracker.res_alloc[i].quota = NULL;
+	}
+	return -ENOMEM;
 }
 
 void mlx4_free_resource_tracker(struct mlx4_dev *dev,
@@ -333,6 +478,14 @@ void mlx4_free_resource_tracker(struct mlx4_dev *dev,
 		}
 
 		if (type != RES_TR_FREE_SLAVES_ONLY) {
+			for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
+				kfree(priv->mfunc.master.res_tracker.res_alloc[i].allocated);
+				priv->mfunc.master.res_tracker.res_alloc[i].allocated = NULL;
+				kfree(priv->mfunc.master.res_tracker.res_alloc[i].guaranteed);
+				priv->mfunc.master.res_tracker.res_alloc[i].guaranteed = NULL;
+				kfree(priv->mfunc.master.res_tracker.res_alloc[i].quota);
+				priv->mfunc.master.res_tracker.res_alloc[i].quota = NULL;
+			}
 			kfree(priv->mfunc.master.res_tracker.slave_list);
 			priv->mfunc.master.res_tracker.slave_list = NULL;
 		}

include/linux/mlx4/device.h

@@ -641,12 +641,23 @@ struct mlx4_counter {
 	__be64	tx_bytes;
 };
 
+struct mlx4_quotas {
+	int qp;
+	int cq;
+	int srq;
+	int mpt;
+	int mtt;
+	int counter;
+	int xrcd;
+};
+
 struct mlx4_dev {
 	struct pci_dev	       *pdev;
 	unsigned long		flags;
 	unsigned long		num_slaves;
 	struct mlx4_caps	caps;
 	struct mlx4_phys_caps	phys_caps;
+	struct mlx4_quotas	quotas;
 	struct radix_tree_root	qp_table_tree;
 	u8			rev_id;
 	char			board_id[MLX4_BOARD_ID_LEN];
@@ -772,6 +783,12 @@ static inline int mlx4_is_master(struct mlx4_dev *dev)
 	return dev->flags & MLX4_FLAG_MASTER;
 }
 
+static inline int mlx4_num_reserved_sqps(struct mlx4_dev *dev)
+{
+	return dev->phys_caps.base_sqpn + 8 +
+		16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev);
+}
+
 static inline int mlx4_is_qp_reserved(struct mlx4_dev *dev, u32 qpn)
 {
 	return (qpn < dev->phys_caps.base_sqpn + 8 +